- commit
- 0b5e614b8d959c143ea7a8d14a0f026d33dcc5e2
- parent
- 206cd6e541a95edabfcc5d3a5529a4eac47a0722
- Author
- Tobias Bengfort <tobias.bengfort@posteo.de>
- Date
- 2023-11-03 09:12
rm shift-heading-level
Diffstat
38 files changed, 221 insertions, 235 deletions
diff --git a/Makefile b/Makefile
@@ -22,7 +22,7 @@ build/%.html: _content/%.html _content/%.yml _templates/*.html build.py 22 22 python build.py _templates/base.html $< _content/$*.yml > $@ 23 23 24 24 _content/%.html: _content/%.md25 -1 pandoc --shift-heading-level-by=1 $< -o $@-1 25 pandoc $< -o $@ 26 26 sed -i 's/<tr class="header">/<tr>/g' $@ 27 27 28 28 build/%.png: _content/%.png
diff --git a/_content/posts/2014-09-09-patch-theory/index.md b/_content/posts/2014-09-09-patch-theory/index.md
@@ -1,5 +1,4 @@1 -1 Introduction2 -1 ============-1 1 ## Introduction 3 2 4 3 Versioning systems have been important tools in software development 5 4 (and other areas) for a long time. In recent years, git has been the @@ -35,8 +34,7 @@ version control for everyone, like a way to eliminate the need for 35 34 manual merges altogether or making git understand that I simply changed 36 35 indentation. 37 3638 -1 The Basics39 -1 ==========-1 37 ## The Basics 40 38 41 39 Say we have a set `S` of *states*. We always use capital letters for 42 40 states. For two states `A, B ∈ S` we say `(A, B)` or shorter `AB` is the @@ -51,8 +49,7 @@ Note that there is some kind of natural composition of changes: When I 51 49 first change from `A` to `B`, and then from `B` to `C`, I have changed 52 50 from `A` to `C` (in formula: `AB * BC = AC`). 53 5154 -1 Patches55 -1 =======-1 52 ## Patches 56 53 57 54 Now imagine you have a repo with some files, one of them called 58 55 "main.py". Then you delete this file. This, in the terms we already @@ -73,8 +70,7 @@ Then again, I would much prefer an axiomatic approach over actually 73 70 defining that relation in every detail. Too strict definitions always 74 71 carry the danger of incompatibility with related theories. 75 7276 -1 The Equivalence Relation77 -1 -------------------------1 73 ### The Equivalence Relation 78 74 79 75 I propose the following axioms for the equivalence relation `~`: 80 76 @@ -107,8 +103,7 @@ based patches, this is only possible if `AC` does not touch any of the 107 103 lines that `CD` changes. And because in the end, both paths arrive at 108 104 `D`, this must mean that `AC` and `CD` are equivalent. 109 105110 -1 Corolarries111 -1 ------------1 106 ### Corolarries 112 107 113 108 - For 1, 2 and 4 there is actually equivalence 114 109 @@ -117,8 +112,7 @@ Corolarries 117 112 - `AA ~ BC <=> B = C` follows from 2 and 4 and shows that there is 118 113 an identity patch we call `id`. 119 114120 -1 Applying a Patch121 -1 -----------------1 115 ### Applying a Patch 122 116 123 117 We can interpret a patch `a` as a function (which is compatible to the 124 118 set theoretic definition of a function) with the following domain and @@ -129,8 +123,7 @@ codomain: 129 123 130 124 We say, a patch `a` is applicable to a state `A`, iff `A ∈ dom(A)`. 131 125132 -1 Composing patches133 -1 =================-1 126 ## Composing patches 134 127 135 128 Up until now we have a notion of what a single patch is. We know how to 136 129 apply it where it is applicable. @@ -164,8 +157,7 @@ defined as `ab := [XZ]`. 164 157 165 158 - It only creates a conflict if `cod(a) ∩ dom(b)` is empty 166 159167 -1 Independence168 -1 -------------1 160 ### Independence 169 161 170 162 Axiom 4 can now be reformulated in terms of patches: 171 163 @@ -174,14 +166,12 @@ Axiom 4 can now be reformulated in terms of patches: 174 166 Patches `a, b ∈ P` for which `ab = ba` is true are called *independent*. 175 167 `id` is independent from all patches. 176 168177 -1 Further Ideas178 -1 =============-1 169 ## Further Ideas 179 170 180 171 These are notes on ideas on some parts that may be interesting to work 181 172 on in the future. 182 173183 -1 Simple Questions184 -1 -----------------1 174 ### Simple Questions 185 175 186 176 There are some questions about patches that seem harmless but are not 187 177 all that easy to get an anser for. Examples: @@ -191,8 +181,7 @@ all that easy to get an anser for. Examples: 191 181 - Is there a patch `a` with `dom(a) ∩ cod(a) != 0` except 192 182 for `id`? 193 183194 -1 Patch sets195 -1 -----------1 184 ### Patch sets 196 185 197 186 The composition I gave in the previous chapter is not associative. 198 187 Example: Say patches `a, b` conflict. Then `(ab)b⁻¹` is not defined, @@ -208,8 +197,7 @@ what the "order that makes most sense" should be and whether it is 208 197 unambiguous. And while most things I said before are compatible with 209 198 common version control systems we use today, this is not. 210 199211 -1 State as Patches212 -1 -----------------1 200 ### State as Patches 213 201 214 202 Now that we have patches we can fix one state we may call `0` and 215 203 identify every state `A ∈ S` with the patch `[0A]` where `0` is @@ -219,10 +207,9 @@ Of course this is only a small subset of patches (more specifically, it 219 207 is `{a ∈ P | 0 ∈ dom(a)}`). But it may have some interesting structural 220 208 properties. 221 209222 -1 Conflicts223 -1 ----------1 210 ### Conflicts 224 211225 -1 ### Group structure-1 212 #### Group structure 226 213 227 214 The set of patches `P` has a structure similar to a group: There is a 228 215 neutral element `id` and for every element `a`, there is an inverse @@ -234,7 +221,7 @@ get a whole lot nicer if we could manage to extend `P` to contain one or 234 221 many elements that represent conflict. I did not come up with a sensible 235 222 way to force `P` into a group, but I guess it would be worth it. 236 223237 -1 ### Manual conflict resultion-1 224 #### Manual conflict resultion 238 225 239 226 An issue that I imagine is very important in developing actual version 240 227 control systems is aiding humans in resolving conflicts. I did only @@ -248,8 +235,7 @@ human in a humane way is something that needs to be done. 248 235 249 236 I did not yet think about it at all. 250 237251 -1 Conclusion252 -1 ==========-1 238 ## Conclusion 253 239 254 240 I believe I have come up with a stable foundation for patch theory. The 255 241 idea to define patches as equivalence classes over the square of states
diff --git a/_content/posts/2015-03-20-resize-partition-on-crypted-disk/index.md b/_content/posts/2015-03-20-resize-partition-on-crypted-disk/index.md
@@ -9,7 +9,7 @@ The trouble is now that the installer chose to use only 10GB for the root 9 9 partition which I used up in the first week. So now I was tasked with resizing 10 10 the partitions. 11 1112 -1 # What I already knew-1 12 ## What I already knew 13 13 14 14 In the past I had used [gparted](http://gparted.org/) for task like this one. 15 15 Unfortunately, gparted does not support encryptet discs. I was very @@ -24,7 +24,7 @@ uses all the rest): 24 24 2. move `home` to the end of the disc 25 25 3. grow `root` 26 2627 -1 # the actual situation-1 27 ## the actual situation 28 28 29 29 The disc layout was roughly like this: 30 30 @@ -41,7 +41,7 @@ simplify my task: 41 41 - logical volumes can use available space anywhere on the physical volume. So 42 42 there is no need to "move" `root`. 43 4344 -1 # what I had to do-1 44 ## what I had to do 45 45 46 46 - The necessary operations can not be applied to mounted volumes. So I had to 47 47 boot into a live system.
diff --git a/_content/posts/2016-03-05-color-contrast/index.md b/_content/posts/2016-03-05-color-contrast/index.md
@@ -12,9 +12,9 @@ personally prefer Sass, so all examples are written in that language. 12 12 One final note: I assume that you have a basic understanding of working with 13 13 colors, especially in CSS. 14 1415 -1 # Solid colors-1 15 ## Solid colors 16 1617 -1 ## Definition-1 17 ### Definition 18 18 19 19 So what is color contrast? There are [many 20 20 definitions](https://en.wikipedia.org/wiki/Color_contrast), but the W3C chose @@ -85,7 +85,7 @@ rationale for the 0.05 is. It prevents the formula from going to infinity 85 85 for near-black colors, but I find that it still produces overly high results 86 86 for those. 87 8788 -1 ## Implementations-1 88 ### Implementations 89 89 90 90 I have seen three types of functions that you may wish to have in your CSS 91 91 preprocessing code: @@ -111,7 +111,7 @@ library called [sass-a11y](https://github.com/at-import/sass-a11y) which 111 111 contains a function of the second type called `a11y-contrast()`. This one seems 112 112 to be correct. 113 113114 -1 # Transparent colors-1 114 ## Transparent colors 115 115 116 116 So now that we know how to calculate color contrast of solid colors, let's turn 117 117 to transparent colors. This topic has been raised by [Lea @@ -153,7 +153,7 @@ I wanted to come up with an algorithm that could easily be implemented in 153 153 existing libraries, so breaking the API was undesirable. So the final sections 154 154 of this article will describe the approach I took. 155 155156 -1 ## Transparent backgrounds-1 156 ### Transparent backgrounds 157 157 158 158 Let's turn to the issue of the unknown backdrop color first. These are some 159 159 approaches I could think of: @@ -215,7 +215,7 @@ in that range. Otherwise, it is the minimum of the white/black cases: 215 215 } 216 216 ``` 217 217218 -1 ## Background/foreground-1 218 ### Background/foreground 219 219 220 220 In the case of solid colors, it was not relevant which of the colors was 221 221 background and which was foreground. Now it is: When a transparent foreground @@ -252,7 +252,7 @@ contrast and its swapped version, I think it is a sensible approach to use a 252 252 } 253 253 ``` 254 254255 -1 ## Implementations-1 255 ### Implementations 256 256 257 257 - Lea Verou created a [tool](https://leaverou.github.io/contrast-ratio/) 258 258 written in JavaScript that reports minimum and maximum possible contrasts @@ -267,7 +267,7 @@ contrast and its swapped version, I think it is a sensible approach to use a 267 267 [sass-planifolia](https://github.com/xi/sass-planifolia) which implements 268 268 the symmetric minimal contrast algorithm. 269 269270 -1 # Conclusion-1 270 ## Conclusion 271 271 272 272 This article covered many details about color contrast as well as a new 273 273 algorithm that supports transparency and can be implemented in existing
diff --git a/_content/posts/2016-06-04-rtl-css/index.md b/_content/posts/2016-06-04-rtl-css/index.md
@@ -1,4 +1,4 @@1 -1 # What is it and why does it concern me?-1 1 ## What is it and why does it concern me? 2 2 3 3 Most western scripts are read from left to right (LTR), so this is what many 4 4 web developers think about when writing CSS. But some scripts, such as Arabic @@ -8,7 +8,7 @@ direction. The complete layout should also be mirrored to match it. 8 8 So say you want to write CSS that can be used with both text directions. What 9 9 do you have to think about? 10 1011 -1 # The problem-1 11 ## The problem 12 12 13 13 HTML and CSS have advanced support for different text directions. You can use 14 14 the [`dir` @@ -22,7 +22,7 @@ The only issue is that `left` and `right` are actually physical directions and 22 22 will not change with text direction, so any `float: left` or `margin-right: 23 23 1em` will not adapt with the text direction. 24 2425 -1 # The solution-1 25 ## The solution 26 26 27 27 Just as relative units such as `em` or `%` are better than absolute `px`, 28 28 relative directions like `start` and `end` are preferable to `left` and @@ -31,7 +31,7 @@ it and `start` / `end` are already available for the [`text-align` 31 31 property](https://developer.mozilla.org/en-US/docs/Web/CSS/text-align) in many 32 32 browsers. 33 3334 -1 # The workaround-1 34 ## The workaround 35 35 36 36 As long as the proper solution is not available, here is a simple way to 37 37 convert a LTR stylesheet to RTL. I am not exactly sure that this is the @@ -65,7 +65,7 @@ complete list, but it should be sufficient: 65 65 In all other cases this is actually worse, apart from being hard to 66 66 maintain. 67 6768 -1 # Outlook: vertical text-1 68 ## Outlook: vertical text 69 69 70 70 CSS3 will take the support for non-latin scripts even further and add support 71 71 for vertical (e.g. top-to-bottom) text. Note however that most vertical scripts @@ -76,7 +76,7 @@ One important thing to remember is that many new keywords from these specs use 76 76 abbreviations such as `vertical-rl` for "vertical-right-left". But you probably 77 77 should not mix that workaround with these new properties anyway. 78 7879 -1 # Further reading-1 79 ## Further reading 80 80 81 81 - <https://github.com/MohammadYounes/rtlcss> 82 82 - <http://www.rtl-this.com/>
diff --git a/_content/posts/2016-12-22-web-is-not-visual/index.md b/_content/posts/2016-12-22-web-is-not-visual/index.md
@@ -3,7 +3,7 @@ the first universally accessible medium. I am not sure if this is actually what 3 3 happened. But a lot of things start to make sense if you look at them from this 4 4 particular perspective. 5 56 -1 # Birth of a new medium-1 6 ## Birth of a new medium 7 7 8 8 Our story begins with an old medium: Print. The invention of the 9 9 printing press around 1440 made books availabe to a lot of people. Scolars like @@ -30,7 +30,7 @@ displays, screen readers or print it on paper. They can be controlled with 30 30 mouse, keyboard, touchscreen or voicecontrol. Whatever a person needs, the 31 31 computer can be programmed to provide the right interface. 32 3233 -1 # The dark ages-1 33 ## The dark ages 34 34 35 35 This could have been a golden age for people with physical or mental 36 36 disabilities who had had trouble with reading before. The new medium finally @@ -47,7 +47,7 @@ This would hava been fine, except that the people who created websites started 47 47 believing it, too. They forgot about semantics and only thought about making 48 48 the content look good in common browsers. 49 4950 -1 # A new hope-1 50 ## A new hope 51 51 52 52 In 2007, finally something happened. People started carrying mobile computers 53 53 with them. These had much smaller screens than traditional computers and were
diff --git a/_content/posts/2017-04-18-ghost-in-the-shell-review/index.md b/_content/posts/2017-04-18-ghost-in-the-shell-review/index.md
@@ -8,7 +8,7 @@ to look at what the movie got right in terms of story. 8 8 **Spoiler alert**: I will talk freely, assuming that you know all there is to 9 9 know about the franchise (or do not care). 10 1011 -1 # The GITS franchise-1 11 ## The GITS franchise 12 12 13 13 There have been [many iterations][3] of ghost in the shell already. The most 14 14 famous one is probably the 1995 movie. I have seen most iterations, but I am @@ -30,7 +30,7 @@ GITS tackles many topics like AI, individuality, gender, violence, and much 30 30 more. There are so many facets to the cyborg idea, and GITs deals with most of 31 31 them. 32 3233 -1 # Donna Haraway-1 33 ## Donna Haraway 34 34 35 35 At some time I stumble across the ["cyborg manifesto"][1] by radical feminist 36 36 Donna Haraway. Haraway's style of writing is somewhat special. She plants ideas @@ -54,7 +54,7 @@ filled with ex-military cyborgs that have to find their place. And sure enough, 54 54 Donna Haraway shows up in the middle of this world as a [forensic doctor in 55 55 Innocence][2]. 56 5657 -1 # The Hollywood movie-1 57 ## The Hollywood movie 58 58 59 59 First of, the new movie copies many scenes and ideas from the previous 60 60 iterations, and it honestly does a great job with that. Imagine how thrilled I @@ -84,7 +84,7 @@ origins in military and capitalism and its revolutionary potential. Sure, the 84 84 movie is not a milestone of intellectuality, but it does a good job of 85 85 introducing this conflict and reminding us that it is real and current. 86 8687 -1 # Conclusion-1 87 ## Conclusion 88 88 89 89 I do not want to argue that this is a great movie. It certainly has many flaws. 90 90 But why does everyone say that it removed the philosophy from the story? Sure,
diff --git a/_content/posts/2017-04-21-math-pairs/index.md b/_content/posts/2017-04-21-math-pairs/index.md
@@ -9,14 +9,14 @@ quality of the algorithm would be: 9 9 - No person should be singled out. 10 10 - It should be simple to explain what everyone has to do. 11 1112 -1 # First approach: Lower bound-1 12 ## First approach: Lower bound 13 13 14 14 In a group of `n` people, there are `n * (n - 1) / 2` pairs in total. At one 15 15 time, there can be at most `floor(n / 2)` pairs. So if `n` is even the minimal 16 16 number of rounds is `n - 1`, if it is odd it is `n`. This makes sense because 17 17 each person has to speak to `n - 1` people. 18 1819 -1 # Second approach: Divide and conquer-1 19 ## Second approach: Divide and conquer 20 20 21 21 The problem can be solved with a simple recursive algorithm: 22 22 @@ -41,7 +41,7 @@ This algorithm is simple, but not optimal. Let's look at the example of 41 41 In the later rounds we have people waiting who could pair up. We should skip 42 42 the first round instead. 43 4344 -1 # Optimal solution: Distances-1 44 ## Optimal solution: Distances 45 45 46 46 Let's first concentrate on the case where `n` is odd. 47 47 @@ -85,7 +85,7 @@ def pairs(n): 85 85 print(r) 86 86 ``` 87 8788 -1 # Conclusion-1 88 ## Conclusion 89 89 90 90 The second algorithm actually is optimal in terms of time. It is also 91 91 relatively easy to understand. The only issue is that, in the even case, person
diff --git a/_content/posts/2017-06-26-simple-language/index.md b/_content/posts/2017-06-26-simple-language/index.md
@@ -3,7 +3,7 @@ of this project was that it would have a "simple German" translation in 3 3 addition to a German one. Little did I know back then that this was the 4 4 beginning of a fantastic journey. 5 56 -1 # The Standards-1 6 ## The Standards 7 7 8 8 First things first. There is a standard that requires us to provide simple 9 9 language alternatives: @@ -43,7 +43,7 @@ helped out with some more tangible proposals ([basiceng][7] and [wpsimple][8]) 43 43 in October; and sometime in December, the ["simple" 44 44 variant subtag][9] was finally accepted. 45 4546 -1 # Summary of the Discussion-1 46 ## Summary of the Discussion 47 47 48 48 You may ask yourself: "Why did it take three months of discussion to add four 49 49 lines to that registry?" There are many reasons for that. @@ -95,7 +95,7 @@ fall back to `de-simple` if the more specific variant was not available. 95 95 > en-US-simple-VoA and en-US-simple-odgenbe and en-US-simple-wp that might 96 96 > work). — [Shawn Steele][12] 97 9798 -1 # Is a new language variant the Answer?-1 98 ## Is a new language variant the Answer? 99 99 100 100 The question remains whether the accessibility issue can be solved on a 101 101 language level alone. Some people argued that complex websites cannot be @@ -122,7 +122,7 @@ to complexity as well: 122 122 > well it is organized (for example using headers, lists, and structures), and 123 123 > how it is presented (font, spacing, width, etc.). — [Shadi Abou-Zahra][22] 124 124125 -1 # Adoption-1 125 ## Adoption 126 126 127 127 More than a year later I came back to see who had adopted this new language 128 128 tag. Unsurprisingly, nobody seemed to have noticed. We had failed to get the @@ -145,7 +145,7 @@ publish simple content already. An obvious choice is [wikipedia][23] (which had 145 145 [ferocious debates][15] about deleting its simple version). But there are 146 146 probably more publishers that might benefit from this. 147 147148 -1 # Conclusion-1 148 ## Conclusion 149 149 150 150 What have we learned? Naming languages is a fascinating and difficult area. 151 151 Simple language is a controversial topic that cannot yet find consensus in the
diff --git a/_content/posts/2017-09-04-rationality/index.md b/_content/posts/2017-09-04-rationality/index.md
@@ -4,7 +4,7 @@ 4 4 > 5 5 > — Donna Haraway, Situated Knowledge 6 67 -1 # Something's cooking-1 7 ## Something's cooking 8 8 9 9 For years I heard about debates between scientists and creationists in the USA. 10 10 I largely ignored those debates because it felt like it was mostly a media @@ -31,7 +31,7 @@ article might not be for you. 31 31 32 32 [^6]: If you understand german, [this articel on Lann Hornscheidt](https://sciencefiles.org/2014/10/31/profxin-lann-hornscheidt-von-der-humboldt-universitat-entfernen/){hreflang=de} is another example. 33 3334 -1 # Empiricism-1 34 ## Empiricism 35 35 36 36 "Science" can be understood in a wider sense, but many people understand it as 37 37 "empirical research". This means that you collect a lot of data and deduce @@ -51,7 +51,7 @@ Imagine you are a brain in a vat that is connected to a computer simulation 51 51 saying that you cannot be certain that this is not the case. Everything you 52 52 sense, all data you collect could be completely unrelated to the real world. 53 5354 -1 # Rationality-1 54 ## Rationality 55 55 56 56 So if you do not trust your senses, the only thing you have left is your 57 57 mind.[^5] Descartes famously took this route and found "I think, therefore I @@ -77,7 +77,7 @@ but we are incapable of thinking that.[^2] 77 77 [^5]: I omit the question what "mind" actually means here. What is the 78 78 difference between "knowing" and "understanding"? Can a machine ever "think"? 79 7980 -1 # Falsification-1 80 ## Falsification 81 81 82 82 So clearly there is no way of *knowing* something. But we can still guess. 83 83 Science ultimately is about making the best guesses possible, or making *less @@ -99,7 +99,7 @@ maybe emotions). 99 99 falsified, "A non-white swan exists" can only be verified. 100 100 [^4]: Politics, not policy! 101 101102 -1 # Choosing a Theory-1 102 ## Choosing a Theory 103 103 104 104 > All models are wrong, but some are useful 105 105 > @@ -119,7 +119,7 @@ Finally, choosing a theory is not a-political: I could believe that jews are 119 119 conspiring to oppress us all. There is probably no data that disproves this 120 120 theory. But I choose to be non-racist. 121 121122 -1 # Conclusion-1 122 ## Conclusion 123 123 124 124 I do see some issues with gender studies, but I also think that their 125 125 consequent refusal of the very concept of objectivity is interesting and
diff --git a/_content/posts/2017-11-05-TDD/index.md b/_content/posts/2017-11-05-TDD/index.md
@@ -10,7 +10,7 @@ I personally feel that TDD is not compatible with my approach to programming. 10 10 I will try to break down that approach by explaining what I don't like about 11 11 TDD. 12 1213 -1 # Design vs Production-1 13 ## Design vs Production 14 14 15 15 For me, programming is about finding the best solution to a problem. It is 16 16 *not* about writing code. These two approaches are often called "design" @@ -40,7 +40,7 @@ If software is designed in this way there is no clear line between spec and 40 40 implementation detail. The code is an immediate and concise representation of a 41 41 mental model. It is its own spec and documentation all at once. 42 4243 -1 # The process of design-1 43 ## The process of design 44 44 45 45 It is hard to pin down how design is actually done. I guess I learned it 46 46 through experience. Rich Hickey did a great explanation of some of the aspects @@ -65,7 +65,7 @@ and ruling out the unlikely ones. It really works best inside of my head. Any 65 65 physical representation like writing or sketching mostly slows me down. The 66 66 occasional scribble can be really helpful though. 67 6768 -1 # Limitations-1 68 ## Limitations 69 69 70 70 Sure, there are some limitations to this approach. You may not always be able 71 71 to express your thoughts in code directly. Maybe the programming language can @@ -82,7 +82,7 @@ and everyone else. Instead, I believe, we should encourage and empower people 82 82 to understand the actual code in the first place, at least to some reasonable 83 83 degree. 84 8485 -1 # What is wrong with TDD?-1 85 ## What is wrong with TDD? 86 86 87 87 > at times I got sucked into that fundamentalist vortex, feeling bad about not 88 88 > following the true gospel. Then I'd try test-first for a few weeks, only to @@ -120,7 +120,7 @@ I totally agree that this is a good trade in some situations. But we need to 120 120 acknowledge that it is actually distracting and harmful in many others. So 121 121 while I think that testing can be beneficial, TDD is a different story. 122 122123 -1 # Conclusion-1 123 ## Conclusion 124 124 125 125 You may have a completely different approach to programming. Maybe TDD gives 126 126 you the structure you need to design and implement great software. Or maybe a
diff --git a/_content/posts/2017-11-20-exponential/index.md b/_content/posts/2017-11-20-exponential/index.md
@@ -2,13 +2,13 @@ How often do you read sentences like "the problem is growing exponentially" or 2 2 "technological progress is exponential"? Do you understand what this should 3 3 tell you? I don't. 4 45 -1 # It's a relation-1 5 ## It's a relation 6 6 7 7 First of all, "exponential" is not a property of a series. It is a *relation*. 8 8 A problem cannot grow exponentially, but it can grow exponentially *in relation 9 9 to* time or people involved or whatever. 10 1011 -1 # It depends on units-1 11 ## It depends on units 12 12 13 13 Much more important, however, is this: "Exponential" depends on the *unit* that 14 14 is used. For example, think about a tone whose frequency rises exponentially in @@ -21,7 +21,7 @@ consumption. If you look at per capita consumption instead, you may well find 21 21 that it has stayed constant. Same for "price of X has grown exponentially": 22 22 Have they taken inflation into account? 23 2324 -1 # It is probably not the right thing-1 24 ## It is probably not the right thing 25 25 26 26 Here is the third issue: Remember when people at your bank tell you that a 27 27 "1.3% interest rate does not seem like much now, but think about what it will @@ -36,7 +36,7 @@ short-term properties to exponential growth, but is vastly different in the 36 36 long term. Again, the long term is what is interesting about exponential 37 37 growth. 38 3839 -1 # Conclusion-1 39 ## Conclusion 40 40 41 41 Maybe I should just accept that "exponential" has become a fancy way to say 42 42 "hugely". That would be sad because exponential growth is a powerful concept if
diff --git a/_content/posts/2017-12-18-personalization/index.md b/_content/posts/2017-12-18-personalization/index.md
@@ -9,7 +9,7 @@ power in the hands of authors and too little power in the hands of users to 9 9 allow effective personalization. Still, I will try to find some practical 10 10 guidelines that may facilitate better personalization on the web. 11 1112 -1 # What exactly is meant by "personalization"-1 12 ## What exactly is meant by "personalization" 13 13 14 14 > Personalization involves tailoring aspects of the user experience to meet the 15 15 > preferences or needs of the user. Technology holds the promise of being @@ -34,7 +34,7 @@ personalization settings. But that would actually be really bad for privacy. 34 34 Instead, personalization should ideally happen completely in the user agent 35 35 without the application even noticing. 36 3637 -1 # Why do we have CSS?-1 37 ## Why do we have CSS? 38 38 39 39 Imagine a world where the rendering of a webpage was completely controlled by 40 40 the browser. In that world, personalization would be easy. It would probably @@ -68,7 +68,7 @@ In the alert case, the additional distinction between success and danger is 68 68 probably implicit from the content. But in other cases, the additional 69 69 semantics may be essential. 70 7071 -1 # Overwriting CSS-1 71 ## Overwriting CSS 72 72 73 73 Now that we are stuck with CSS, the only option to enable personalization is to 74 74 overwrite it. This is also why the [User Agent Accessibility @@ -108,7 +108,7 @@ removed support for user level 108 108 stylesheets](https://bugs.chromium.org/p/chromium/issues/detail?id=347016) in 109 109 2014. 110 110111 -1 # Conclusion-1 111 ## Conclusion 112 112 113 113 The web is the most accessible medium we have ever had. But still, it is far 114 114 from perfect. Semantic HTML enables us to use a wide range of representations –
diff --git a/_content/posts/2017-12-30-icons/index.md b/_content/posts/2017-12-30-icons/index.md
@@ -6,7 +6,7 @@ The W3C's [Cognitive Accessibility Roadmap and Gap 6 6 Analysis](https://www.w3.org/TR/coga-gap-analysis/) proposes to allow users to 7 7 personalize icons. This is my proposal on how that could be done. 8 89 -1 # A brief history of icons on the web-1 9 ## A brief history of icons on the web 10 10 11 11 In order to know how a native HTML `<icon>` element could work, we should look 12 12 at existing solutions, as well as their problems and benefits. @@ -37,7 +37,7 @@ SVG allows to style multi-colored icons with CSS. However, knowledge of image 37 37 internals is required for that. The accessibility with these is better, but 38 38 still far from perfect. 39 3940 -1 # Standalone vs. Presentational-1 40 ## Standalone vs. Presentational 41 41 42 42 In my experience, there are two distinct usecases for icons: Sometimes icons 43 43 are used with a label. Icons like this are not strictly necessary, but they @@ -77,7 +77,7 @@ The HTML that is currently required for this is a bit redundant: 77 77 </button> 78 78 ``` 79 7980 -1 # My proposal-1 80 ## My proposal 81 81 82 82 I propose to add something like this to HTML: 83 83 @@ -107,7 +107,7 @@ There is just one restriction for privacy reasons: Web pages should not be able 107 107 to find out which users personalize their icons. For that reason, all icons 108 108 must be square. 109 109110 -1 # Open questions-1 110 ## Open questions 111 111 112 112 - How does branding/personalization work exactly? It somewhat works with fonts, 113 113 but that's still far from perfect. @@ -119,7 +119,7 @@ must be square. 119 119 can new icons be added to the registry without the hassle of a full 120 120 standardization process? 121 121122 -1 # Conclusion-1 122 ## Conclusion 123 123 124 124 Icons are used literally everywhere. Adding them to HTML itself would allow to 125 125 make life easier for web developers at the same time as improving
diff --git a/_content/posts/2018-01-14-consensus-meetings/index.md b/_content/posts/2018-01-14-consensus-meetings/index.md
@@ -6,7 +6,7 @@ There are meetings that are not ineffective. I have been in some. Those were 6 6 the ones where everyone involved knew some rules. These rules are not black 7 7 magic, but they are also not all that easy to explain. Still, I will try: 8 89 -1 # Value the social aspects-1 9 ## Value the social aspects 10 10 11 11 A meeting of the kind I am talking about mainly exists to make decisions. But 12 12 it is also one of the rare occasions where your team comes together. It is a @@ -22,12 +22,12 @@ decision as fast as possible?" But I think you should allow yourself to get 22 22 side-tracked sometimes. Get to know each other and grow as a team. This will 23 23 greatly improve things in the long run. 24 2425 -1 # Stick to the topic-1 25 ## Stick to the topic 26 26 27 27 This one is obvious and I will not waste too much time with it: Contrary to what 28 28 I said before, you should of course always try to stick to the topic. 29 2930 -1 # Decide by consensus-1 30 ## Decide by consensus 31 31 32 32 Before I had experience with consensus-based decisions, I thought it was a 33 33 grossly ineffective method. It turns out the exact opposite is true. The @@ -51,7 +51,7 @@ Simple as that. 51 51 A more detailed description of the process can be found in 52 52 [RFC7282](https://tools.ietf.org/html/rfc7282). 53 5354 -1 # Be aware of common fallacies-1 54 ## Be aware of common fallacies 55 55 56 56 The biggest issue with consensus-based decisions is that they are susceptible 57 57 to social forces. You should be very critical of your position in the team. If @@ -73,7 +73,7 @@ decision, i.e. leave everything as it currently is. Maybe this is the option 73 73 you all can agree on. Or maybe it is bad enough for everyone to compromise on 74 74 something else. 75 7576 -1 # Just do it-1 76 ## Just do it 77 77 78 78 I honestly believe that if you follow these few rules you will have great 79 79 meetings. Of course, this requires that everyone involved knows about them.
diff --git a/_content/posts/2018-05-28-zen/index.md b/_content/posts/2018-05-28-zen/index.md
@@ -16,7 +16,7 @@ pretense to nerd about philosophy for hundreds of pages. But where Sophie 16 16 explores the history of western philosophy, Pirsig mostly writes about his own 17 17 ideas. 18 1819 -1 # Content-1 19 ## Content 20 20 21 21 > The study of the art of motorcycle maintenance is really a miniature study of 22 22 > the art of rationality itself. Working on a motorcycle, working well, caring, @@ -57,7 +57,7 @@ by Donne Haraway and especially [agential 57 57 cuts](https://en.wikipedia.org/wiki/Karen_Barad#Agential_realism) by Karen 58 58 Barad. 59 5960 -1 # Conclusion-1 60 ## Conclusion 61 61 62 62 > The real cycle you're working on is a cycle called yourself. The machine that 63 63 > appears to be "out there" and the person that appears to be "in here" are not
diff --git a/_content/posts/2018-10-28-aria-quirks/index.md b/_content/posts/2018-10-28-aria-quirks/index.md
@@ -12,7 +12,7 @@ opinion on any of these, I would love to hear from you! 12 12 13 13 [1]: http://getbootstrap.com/docs/4.1/components/ 14 1415 -1 # Building my own assistive technology-1 15 ## Building my own assistive technology 16 16 17 17 Most screen readers let you navigate along the headings, links, or landmarks in 18 18 a page. Since visual scanning is impossible for most visually impaired users, @@ -34,7 +34,7 @@ any bugs, please make sure to [report them][3]. 34 34 [2]: https://addons.mozilla.org/de/firefox/addon/a11y-outline/ 35 35 [3]: https://github.com/xi/a11y-outline/issues 36 3637 -1 # ARIA primer-1 37 ## ARIA primer 38 38 39 39 So, what is this ARIA thing? 40 40 @@ -66,12 +66,12 @@ operating systems. The layers in between are defined in different specs: 66 66 - The mapping between ARIA and OS APIs is defined in the [Core Accessibility API Mappings (Core-AAM)](https://www.w3.org/TR/core-aam-1.1/). 67 67 - Some additional parts have been split out, e.g. the [Accessible Name and Description Computation](https://www.w3.org/TR/accname-1.1/). 68 6869 -1 # Quirks-1 69 ## Quirks 70 70 71 71 Now with the introduction out of the way, let's get to the weird stuff. We 72 72 start slow though. 73 7374 -1 ## Landmarks-1 74 ### Landmarks 75 75 76 76 Landmark is a category of roles. Here is a typical layout for a page: 77 77 @@ -126,7 +126,7 @@ Forms, on the other hand, only become landmarks if they have an explicit label: 126 126 </form> 127 127 ``` 128 128129 -1 ## Accessible name calculation-1 129 ### Accessible name calculation 130 130 131 131 Each element has a name and description. The algorithem to calculate them is 132 132 crazy complicated as it contains recursion and tons of special cases. My own @@ -218,7 +218,7 @@ If I follow the spec to the word the name for the select element should be 218 218 sure this is not intended. Probably only the selected option should be 219 219 included. 220 220221 -1 ## CSS is kinda relevant-1 221 ### CSS is kinda relevant 222 222 223 223 Ok, this is web 1x1, right? HTML is for content, CSS is for presentation, JS is 224 224 for behavior. So the ARIA semantics should surely not be influenced by CSS, @@ -269,7 +269,7 @@ when calculating names. 269 269 <p><em>un</em>subscribe</p> <!-- unsubscribe --> 270 270 ``` 271 271272 -1 ## HTML-AAM overrides other specs-1 272 ### HTML-AAM overrides other specs 273 273 274 274 Remember how I wrote about this nice hierarchy in the beginning? It was HTML - 275 275 HTML-AAM - ARIA - Core-AAM - OS APIs. Well …, that is not actually true. @@ -287,7 +287,7 @@ are completely different than the default algorithm. In this case, the 287 287 otherwise rarely used `title` attribute wins. The `hidden` flag is simply 288 288 ignored. 289 289290 -1 ## Conflicts-1 290 ### Conflicts 291 291 292 292 Ready for the final quirk? In my opionion this is the worst one and also the 293 293 one that is probably the hardest to fix. @@ -309,7 +309,7 @@ documented exceptions: 309 309 > same implicit semantic. 310 310 > — <https://www.w3.org/TR/core-aam-1.1/#mapping_conflicts> 311 311312 -1 # Conclusion-1 312 ## Conclusion 313 313 314 314 ARIA contains some unexpected bits and even some bugs that result from high 315 315 complexity. Maybe this complexity could be reduced in some places.
diff --git a/_content/posts/2019-01-13-visual/index.md b/_content/posts/2019-01-13-visual/index.md
@@ -7,7 +7,7 @@ For some reason, we treat sight special. We say "I see" when we really mean "I 7 7 understand". [Bret Victor](http://worrydream.com/#!/LearnableProgramming) even 8 8 claims that "People understand what they can see". Why is that? 9 910 -1 # Senses-1 10 ## Senses 11 11 12 12 First, it is important to note that sight is especially well developed in 13 13 humans compared to other mammals. Dogs for example have an advanced sense of @@ -33,7 +33,7 @@ required to pass them on orally from generation to generation. Writing them 33 33 down was ultimately the better way to record information. (Today, we can also 34 34 record sounds. But that is a comparatively recent invention.) 35 3536 -1 # Text-1 36 ## Text 37 37 38 38 Maybe comparing sight to other senses is not the right approach: I think that 39 39 what Bret Victor enjoys about seeing a diagram is that it is more intuitive to @@ -57,7 +57,7 @@ trends are worrisome. Today it is hard to find a complicated text that is not 57 57 "lightened up" by some image. I fear that these graphics are intuitive to a 58 58 point where you *believe* to understand more than actually understanding. 59 5960 -1 # Conclusion-1 60 ## Conclusion 61 61 62 62 After reading a lot about the topic I now have a better understanding of why 63 63 visuals are so important in web design. But I still think that we should be
diff --git a/_content/posts/2019-05-02-debian-packages/index.md b/_content/posts/2019-05-02-debian-packages/index.md
@@ -4,7 +4,7 @@ passes all lintian checks and could be included in the official repositories. 4 4 If you just want to use the package for yourself, the process is not actually 5 5 that hard. 6 67 -1 # Why should I create a debian package?-1 7 ## Why should I create a debian package? 8 8 9 9 If you are using debian or any of its derivatives such as ubuntu or mint, you 10 10 typically install software on your system as debian packages using the `apt` @@ -26,7 +26,7 @@ knows about both the files and dependencies and can remove them automatically. 26 26 As an added benefit, you get an automatic update if your distribution starts 27 27 packaging the software. 28 2829 -1 # The joy of meta-packages-1 29 ## The joy of meta-packages 30 30 31 31 A special kind of package you may want to create are meta-packages. A 32 32 meta-package is one that only defines dependencies and contains no files of its @@ -39,7 +39,7 @@ everything else gets installed automatically. That way, the list of manually 39 39 installed packages stays short and concise. Meta-packages are also great to 40 40 manage build-dependencies for other packages. 41 4142 -1 # How to create a package-1 42 ## How to create a package 43 43 44 44 So here is the tutorial: 45 45 @@ -63,7 +63,7 @@ Simple as that. 63 63 64 64 You can install the package using `sudo apt install {filename}`. 65 6566 -1 # Taking the next step-1 66 ## Taking the next step 67 67 68 68 Now that you have a simple package, there are some things you could look into 69 69 to improve it: @@ -78,7 +78,7 @@ to improve it: 78 78 - Instead of copying the files manually into the package folder, instruct `make 79 79 install` to do it automatically. 80 8081 -1 # Conclusion-1 81 ## Conclusion 82 82 83 83 Creating debian packages is not so hard after all. It may not be as refined as 84 84 the [`PKGBUILD` system in
diff --git a/_content/posts/2019-06-24-terminal-colors/index.md b/_content/posts/2019-06-24-terminal-colors/index.md
@@ -5,7 +5,7 @@ this: 5 5 6 6 I tried to find out why that happens. This article is a summary of what I found. 7 78 -1 # ANSI colors-1 8 ## ANSI colors 9 9 10 10 Modern terminals can display 256 colors. But when talking about terminal color 11 11 schemes, we talk about a set of 16 colors, known as ANSI colors. @@ -29,7 +29,7 @@ bother to define which colors should be used together. 29 29 30 30 So let's see how popular terminal themes handle this. 31 3132 -1 # Tango-1 32 ## Tango 33 33 34 34  35 35 @@ -46,7 +46,7 @@ I am used to this theme and would like to stick with it. It is listed here as 46 46 some kind of baseline. Later in the article, I will try to improve it using 47 47 concepts from other themes. 48 4849 -1 # Solarized-1 49 ## Solarized 50 50 51 51  52 52 @@ -89,7 +89,7 @@ of the bright colors with greys. 89 89 So solarized is a mixed bag. I like its structured approach but don't agree 90 90 with most of the design decisions. 91 9192 -1 # Gruvbox-1 92 ## Gruvbox 93 93 94 94  95 95 @@ -126,7 +126,7 @@ the "black" color as background and the "bright white" color as foreground. 126 126 That is not the case for solarized and yet another reason why application 127 127 authors cannot rely on these colors. 128 128129 -1 # Tango with consistent lightness-1 129 ## Tango with consistent lightness 130 130 131 131 Luckily, I am not trying to build a light theme, so many of the issues 132 132 mentioned above don't directly affect me. But still, the only tangible advice I @@ -167,7 +167,7 @@ adjusting the bright colors: 167 167 168 168  169 169170 -1 # Conclusion-1 170 ## Conclusion 171 171 172 172 Honestly, I am not sure I found out anything. But I will try to cobble together 173 173 some advice: @@ -193,7 +193,7 @@ as motivation for this whole thing? After taking it I realized that I had 193 193 forgotten to switch back to the original tango theme. So it was taken using the 194 194 already improved version. That's how little has improved. 195 195196 -1 # One more fun idea-1 196 ## One more fun idea 197 197 198 198 This is an idea I had that didn't really fit into the article: 199 199
diff --git a/_content/posts/2020-01-20-open-source/index.md b/_content/posts/2020-01-20-open-source/index.md
@@ -23,7 +23,7 @@ of different types of software projects. The categorization that I want to 23 23 propose is about whether a software is meant to be generic and who its target 24 24 audience is. 25 2526 -1 # Side projects-1 26 ## Side projects 27 27 28 28 On the non-generic technical-user side of the spectrum are hobby projects. 29 29 Things you do by yourself and for yourself. You get payed in fun and experience @@ -36,7 +36,7 @@ it useful. I personally really like to look at existing attempts of solving a 36 36 problem. I mostly don't end up using any of these projects, but I learn a lot 37 37 from them anyways. 38 3839 -1 # Bespoke applications-1 39 ## Bespoke applications 40 40 41 41 Companies need software for their internal processes. This software must be 42 42 usable for non-technical users. Developers have no intrinsic motivation to @@ -49,7 +49,7 @@ else might find it useful and maybe even contribute back some bug fixes. In 49 49 addition, using GPL allows you to use GPL libraries. Choosing open source also 50 50 tends to help with recruiting developers. 51 5152 -1 # Infrastructure-1 52 ## Infrastructure 53 53 54 54 With the first two categories, publishing the code was a nice addition, but the 55 55 software was not really meant to be used by anyone else. While these are @@ -79,7 +79,7 @@ important a piece of infrastructure is, the more important it is that there is 79 79 a clear governance model. If the US government provided the funding for 80 80 openssl, would that have any influence on NSA backdoors? 81 8182 -1 # End-user applications-1 82 ## End-user applications 83 83 84 84 Finally, on the generic end-user end of the spectrum we have projects 85 85 such as firefox, libre office, gnome, or wordpress. @@ -94,7 +94,7 @@ I honestly don't know where projects from this category get their funding from. 94 94 I can just imagine that it is not simple. All the more reason to be impressed 95 95 by the projects that do exist. 96 9697 -1 # Proprietary software-1 97 ## Proprietary software 98 98 99 99 I talked a lot about the different categories of open source and their 100 100 different benefits. But what about proprietary software? @@ -115,7 +115,7 @@ there are semen-free cookies available for free. 115 115 I understand that other people will associate secrecy and a high price with 116 116 high quality. So this is totally subjective. But to me it feels shady. 117 117118 -1 # Price of work-1 118 ## Price of work 119 119 120 120 Mark Tarver criticises that there is a culture of self-exploitation in open 121 121 source, where people put in much more than they get out. This is definitely not @@ -134,7 +134,7 @@ ways to fund small infrastructure projects is definitely helpful, the other 134 134 approach to establish clear boundaries and go back to being a side project is 135 135 equally valid. 136 136137 -1 # Conclusion-1 137 ## Conclusion 138 138 139 139 I think the open source community is in a really great spot: We publish many 140 140 side projects to learn from one another, we get paid to build bespoke
diff --git a/_content/posts/2020-05-16-pad/index.md b/_content/posts/2020-05-16-pad/index.md
@@ -38,7 +38,7 @@ that would otherwise be unintelligible. 38 38 39 39 Long story short: I built my own plain-text pad. 40 4041 -1 # The Goal-1 41 ## The Goal 42 42 43 43 The guiding principle had to be simplicity: I wanted to get 80% of 44 44 functionality for 20% effort. @@ -51,7 +51,7 @@ To take this further I was also willing to cut corners in terms of 51 51 collaboration: My assumption was that concurrent edits in the same place are 52 52 rare and that there is really no perfect way to handle them anyway. 53 5354 -1 # Implementation-1 54 ## Implementation 55 55 56 56 I read the original [etherpad design 57 57 document](https://github.com/ether/etherpad-lite/raw/master/doc/easysync/easysync-full-description.pdf) @@ -105,7 +105,7 @@ change on conflict. As this happens quickly, users can react and re-apply their 105 105 change manually. The big lesson here is that convergence is more important than 106 106 the quality of the merge. 107 107108 -1 # Testing-1 108 ## Testing 109 109 110 110 I created bots for testing that helped me find some bugs. But since many of my 111 111 simplifications relied on assumptions about real-world usage I had to test with @@ -115,7 +115,7 @@ When I was confident enough I asked some coworkers if we could use my pad 115 115 for a real meeting. And what can I say. During the meeting I completely forgot 116 116 that this was my new tool. It just worked. 117 117118 -1 # Conclusion-1 118 ## Conclusion 119 119 120 120 So there it is. You can find the code at <https://git.ce9e.org/pad/> and a demo 121 121 at <https://pad.ce9e.org/>. It ended up being ~220 lines of javascript. That is
diff --git a/_content/posts/2020-10-10-thunderbird/index.md b/_content/posts/2020-10-10-thunderbird/index.md
@@ -12,7 +12,7 @@ work!" 12 12 I don't want to feel like it was all for nothing, at least I want to write 13 13 down what I learned. So this is me rambling about some random software release. 14 1415 -1 # WebExtensions-1 15 ## WebExtensions 16 16 17 17 So let's start at the beginning. What was this release all about? If you look 18 18 through the [official release notes][1] it sound like the focus was on UI @@ -29,7 +29,7 @@ completely, and some may not be possible with the new APIs at all. But in 29 29 contrast to the UI changes described above I can see a lot of value in the 30 30 refactoring work that is being done, even if it causes me some trouble. 31 3132 -1 # Builtin PGP support-1 32 ## Builtin PGP support 33 33 34 34 One of the most popular extensions that was not going to be supported was 35 35 [Enigmail][3]. The Thunderbird developers of course would never dare to release @@ -85,7 +85,7 @@ You can also vote for this UI on the [feature request][8] I created. 85 85 (Full disclosure: I am not a genius designer. I just copied what enigmal had 86 86 done before.) 87 8788 -1 # My own extension-1 88 ## My own extension 89 89 90 90 All that was the easy part, the part where I can simply vent about bad design 91 91 without any responsibility of my own. This part now is about an extension I @@ -124,7 +124,7 @@ So the big question is what will happen to Experiments in the future: Is this 124 124 just a temporary playground to test different interfaces until we settle on a 125 125 common set? Or will it be around indefinitely to support fringe use cases? 126 126127 -1 # Conclusion-1 127 ## Conclusion 128 128 129 129 This is not really about bad design because I think with all the refactoring 130 130 and UI cleanup as well as the built-in PGP support Thunderbird 78 is the best
diff --git a/_content/posts/2020-11-21-file-type-101/index.md b/_content/posts/2020-11-21-file-type-101/index.md
@@ -19,7 +19,7 @@ file types. So this is what this article is about. After reading this you will 19 19 have a basic understanding of what any file is and how you might be able to 20 20 interact with it. 21 2122 -1 # A file is not a program-1 22 ## A file is not a program 23 23 24 24 Before I get into it, I have to clear up a common misconception: 25 25 @@ -39,7 +39,7 @@ files. 39 39 40 40 With that out of the way, let's get to the actual files. 41 4142 -1 # Text files-1 42 ## Text files 43 43 44 44 You probably already know that files just consist of ones and zeroes, also 45 45 called *bits*. How these bits are interpreted depends on the specific type, @@ -70,7 +70,7 @@ If you come across a file which you don't know it is often a good idea to look 70 70 at it in a text editor. If the file happens to contain text, you can read it 71 71 and maybe understand enough to know what to do next. 72 7273 -1 # XML-1 73 ## XML 74 74 75 75 Mapping bits to characters already gives us some structure, but apparently not 76 76 enough. So people have invented different formats on top of that. Probably the @@ -96,7 +96,7 @@ brackets. 96 96 XML is used virtually everywhere. For example: Every website is essentially 97 97 just an XML file. 98 9899 -1 # ZIP-1 99 ## ZIP 100 100 101 101 Text files are great because they are much easier to read compared to a stream 102 102 of ones and zeroes. However, that comes at the price of being less efficient. @@ -113,7 +113,7 @@ This is actually how many file types work these days. For example, try changing 113 113 the file extension of any MS Office file to `.zip`. You can now unpack it and 114 114 see what it contains: You guessed it, a bunch of XML files! 115 115116 -1 # Conclusion-1 116 ## Conclusion 117 117 118 118 Of course there are a lot more file types. But text files and ZIP alone already 119 119 cover a lot of ground.
diff --git a/_content/posts/2020-12-28-wayland-wm/index.md b/_content/posts/2020-12-28-wayland-wm/index.md
@@ -2,7 +2,7 @@ If a voice in your had is currently screaming "THERE ARE NO WINDOW MANAGERS IN 2 2 WAYLAND, THEY ARE CALLED COMPOSITORS" you swallowed the bait. That is exactly 3 3 what I want to write about. But let me start at the beginning: 4 45 -1 # A short history of wayland-1 5 ## A short history of wayland 6 6 7 7 For the last 30-odd years graphics on Unix were dominated by X11. In 2008 the 8 8 people who maintained X11 declared that for the good of the project, starting @@ -14,7 +14,7 @@ was looking good. Fedora uses wayland by default since 2016. But most other 14 14 distros in 2020 still use an X11 server, which at this point is [largly 15 15 unmaintained](https://news.ycombinator.com/item?id=24884988). What went wrong? 16 1617 -1 # The modular X11 desktop-1 17 ## The modular X11 desktop 18 18 19 19 X11 was about painting pixels to the screen. But it also provided APIs 20 20 like [EWMH](https://specifications.freedesktop.org/wm-spec/wm-spec-latest.html) @@ -39,7 +39,7 @@ standardized APIs. If you want to port your desktop to wayland, you have to 39 39 replace everything all at once. That might be possible for Gnome and KDE, but 40 40 not for the ecosystem of modular components. 41 4142 -1 # The wlroots project-1 42 ## The wlroots project 43 43 44 44 This decision by the wayland maintainers did not neccesarily mean the end of 45 45 the modular desktop though. Someone else could step in and define the necessary @@ -56,7 +56,7 @@ repo, but there is little progress. And even though there are many 56 56 wlroots-based compositors, the only one ready for production is 57 57 [sway](https://github.com/swaywm/sway). 58 5859 -1 # Sway-1 59 ## Sway 60 60 61 61 About once a year I look at wayland, mess around a little, get frustrated, and 62 62 soon decide that I should wait another year. This year around I tried sway, @@ -79,7 +79,7 @@ but none of these is really in a usable state. I looked at their code and 79 79 quickly decided that writing one myself was also out of the question. So what 80 80 could I do? 81 8182 -1 # Writing a wayland window manager-1 82 ## Writing a wayland window manager 83 83 84 84 I had already created a toy X11 window manager (based on 85 85 [dwm](https://dwm.suckless.org/)) in the past. It is not actually that hard: @@ -187,7 +187,7 @@ if __name__ == '__main__': 187 187 con.main() 188 188 ``` 189 189190 -1 # Conclusion-1 190 ## Conclusion 191 191 192 192 Wayland is the future and outside of the big desktop environments, sway is the 193 193 only viable option right now. So if, like me, you want to use wayland but don't
diff --git a/_content/posts/2021-04-03-zero-cost-optimizations/index.md b/_content/posts/2021-04-03-zero-cost-optimizations/index.md
@@ -5,7 +5,7 @@ practically zero. So why not do them? 5 5 In this post I want to share some thoughts about the structure of this kind 6 6 of optimization. 7 78 -1 # Examples-1 8 ## Examples 9 9 10 10 That sounds very abstract, so what do I mean by this? 11 11 @@ -23,12 +23,12 @@ One example from another area of life would be veganism. For the largest part 23 23 it is healthier, kinder, and more sustainable than eating meat. There is no 24 24 extra effort involved. I see no reason against it. 25 2526 -1 # Counter arguments-1 26 ## Counter arguments 27 27 28 28 Obviously not all people value accessibility, code styling and veganism. I want 29 29 to look at some common counter arguments: 30 3031 -1 ## You ain't gonna need it-1 31 ### You ain't gonna need it 32 32 33 33 [YAGNI](http://c2.com/xp/YouArentGonnaNeedIt.html) is a reminder against 34 34 feature creep. Even if you think you will need a feature, don't implement it @@ -37,7 +37,7 @@ situations. However, all examples I mentioned were about doing a thing 37 37 *differently* rather than adding something new on top. So I don't think it 38 38 applies. 39 3940 -1 ## Effective altruism-1 40 ### Effective altruism 41 41 42 42 [Effective 43 43 altruism](https://www.jefftk.com/p/effective-altruism-and-everyday-decisions) @@ -51,7 +51,7 @@ Still, I see no reason not to do them. I would rephrase this critizism as: 51 51 "Don't spend time arguing about it, just do it". (I am aware of the irony of me 52 52 writing a blog post about this topic. We all have our faults.) 53 5354 -1 ## Virtue signalling-1 54 ### Virtue signalling 55 55 56 56 [Virtue signalling](https://www.urbandictionary.com/define.php?term=Virtue%20Signalling) 57 57 is the allegation that people do zero cost optimizations not for their benefits @@ -66,7 +66,7 @@ not communicate"](https://en.wikipedia.org/wiki/Paul_Watzlawick). I would argue 66 66 that this form of communication is even useful in most cases. It is good to 67 67 know whether the person I talk to actually knows and cares about the topic. 68 6869 -1 # Habits and teams-1 69 ## Habits and teams 70 70 71 71 When I say these optimizations come at zero cost, that is only true for the 72 72 long term. When you are just starting you may need to change your habits, and @@ -81,7 +81,7 @@ As you usually communicate with different people, you need to have some degree 81 81 of flexibility in your habits. If you over-optimize, you will have a hard time 82 82 fitting in. 83 8384 -1 # Conclusion-1 84 ## Conclusion 85 85 86 86 Zero cost optimizations, also called "virtue signalling", are efficient but 87 87 ineffective actions based on habits. Arguing about them is not worth the time.
diff --git a/_content/posts/2021-08-20-linux-service-architecture/index.md b/_content/posts/2021-08-20-linux-service-architecture/index.md
@@ -8,7 +8,7 @@ Maybe I am totally mistaken and what I want to point out is clear as daylight 8 8 to everyone but me. But I have never seen it spelled out. So I will attempt to 9 9 do the spelling. 10 1011 -1 # Chapter 1: Traditional Unix-1 11 ## Chapter 1: Traditional Unix 12 12 13 13 A process on a unix system traditionally acts as a proxy for the user. If I use 14 14 some software, that software can do anything that I can do. And if I use some @@ -18,7 +18,7 @@ There have been attempts at more fine-grained access control with systems such 18 18 as SELinux or AppArmor, but writing correct configuration for them is tedious. 19 19 So most processes run unconfined. 20 2021 -1 # Chapter 2: The rise of system services-1 21 ## Chapter 2: The rise of system services 22 22 23 23 Running software as root is dangerous, so we want to avoid that. Additionally, 24 24 especially on single-user systems, it does not make much sense to require root @@ -49,7 +49,7 @@ would cause many heated discussion throughout the next years. 49 49 In today's linux there are few reasons to ever use `sudo`. (At least in theory, 50 50 because I for one still use `mount` over `udiskctl` and `apt` over `pkcon`.) 51 5152 -1 # Chapter 3: Same concept, different name-1 52 ## Chapter 3: Same concept, different name 53 53 54 54 After the threat of software running as root was banned, people quickly 55 55 realized that running as regular users also comes with risks. A tame-looking @@ -90,7 +90,7 @@ Running untrusted code feels much less scary when it is sandboxed and 90 90 applications need stable API to program against, so this whole idea is closely 91 91 related to that of portals. 92 9293 -1 ## Chapter 4: What it all means-1 93 ### Chapter 4: What it all means 94 94 95 95 I think I have now mentioned every single hot linux topic from the past 15 96 96 years at least once. And everything is connected!
diff --git a/_content/posts/2022-02-20-minimalism/index.md b/_content/posts/2022-02-20-minimalism/index.md
@@ -2,7 +2,7 @@ Ok, this is going to be a rambling mess. There are some things I want to write 2 2 down and I don't yet know how to put the different threads together. So let's 3 3 untangle that mess. 4 45 -1 # I am not yet sure where this is going-1 5 ## I am not yet sure where this is going 6 6 7 7 The first thread is about the linux desktop community. Compared to others, this 8 8 community is tiny. Shell scripting for example is a much bigger community @@ -23,7 +23,7 @@ just a function of the size of the community. On any given web topic you will 23 23 likely find 10 helpful blog posts. For linux desktop stuff you are lucky if you 24 24 find anything relevant at all. 25 2526 -1 # A lot of quotes pushed into a single section-1 26 ## A lot of quotes pushed into a single section 27 27 28 28 Which brings us to the second thread: Why should anyone even want to build 29 29 their own desktop? Why not stick with gnome or KDE or whathaveyou? @@ -56,7 +56,7 @@ an important part of the journey. 56 56 > together. 57 57 > — Robert M. Pirsig, Zen and the Art of Motorcycle Maintenance 58 5859 -1 # Information walks into a bar-1 59 ## Information walks into a bar 60 60 61 61 The third thread is about displaying up-to-date information in a status bar. In 62 62 order for it to really be live, we need to push new information into the UI as @@ -75,7 +75,7 @@ unicode symbols to get some more interesting graphics. With 75 75 [i3blocks](https://github.com/vivien/i3blocks) and similar projects there is 76 76 even an ecosystem of modular tools that can generate input for these bars. 77 7778 -1 # How did I get here?-1 78 ## How did I get here? 79 79 80 80 The next thread is about the people in this community. And this is where it 81 81 gets weird. @@ -111,7 +111,7 @@ vim and dwm and how the jews are using mainstream media to brainwash us. 111 111 112 112 So yeah, this community is full of racists. 113 113114 -1 # Let's quickly talk about something else-1 114 ## Let's quickly talk about something else 115 115 116 116 While using pipes is obviously the much cleaner option, most mainstream 117 117 applications broadcast their status via dbus. So I did what seemed obvious and @@ -121,7 +121,7 @@ display those. 121 121 122 122 The result is available on github: <https://github.com/xi/polybar-status-indicators> 123 123124 -1 # I have no clue how to end this-1 124 ## I have no clue how to end this 125 125 126 126 I really really like minimalist software. I seem to share this 127 127 longing for purity with fascists. That's fucked up.
diff --git a/_content/posts/2022-06-25-hard-work/index.md b/_content/posts/2022-06-25-hard-work/index.md
@@ -12,7 +12,7 @@ David R. MacIver writes: 12 12 I found that distinction interesting to reflect about my work, so I will try to 13 13 add some of my thoughts to the discussion. 14 1415 -1 # The problem with automation-1 15 ## The problem with automation 16 16 17 17 My job consists mostly of identifying areas where people in the organisation 18 18 have to do hard work and finding better ways to do that. @@ -51,7 +51,7 @@ best choice in every case. While reflecting on this, I realize that I sometimes 51 51 focus too much on eliminating hard work and do not actually consider the 52 52 downsides of the alternatives. 53 5354 -1 # A better option-1 54 ## A better option 55 55 56 56 Now you may be asking: "Why are there even numbers on paper?" And I think that 57 57 is exactly the right question to ask here. Maybe we can just talk to the other
diff --git a/_content/posts/2022-09-10-contrast-algorithms/index.md b/_content/posts/2022-09-10-contrast-algorithms/index.md
@@ -6,7 +6,7 @@ I don't want to pick on the color.js authors specifically. They are generally 6 6 doing an awesome job. This documentation just happens to aggregate many of the 7 7 claims that can be found throughout the internet. 8 89 -1 # The many faces of Weber contrast-1 9 ## The many faces of Weber contrast 10 10 11 11 Let's start with some fractions. 12 12 @@ -55,7 +55,7 @@ Admittedly, these formulas are not exactly the same. But I cannot see how you 55 55 can say that simple contrast is unfit for real-world use and not say something 56 56 similar about the others. 57 5758 -1 # Equivalence of contrast formulas-1 58 ## Equivalence of contrast formulas 59 59 60 60 I want to try to formalize this notion of "similarity" between different 61 61 formulas. @@ -83,7 +83,7 @@ function would have to be linear and Weber would not be equivalent to Simple 83 83 contrast. I do think this stricter definition has merit in some cases, but I 84 84 will work with the weaker definition in this post. 85 8586 -1 # Lightness difference-1 86 ## Lightness difference 87 87 88 88 The authors also mention lightness difference `Lmax - 89 89 Lmin` as a way to calculate contrast: @@ -110,12 +110,12 @@ claims that the other contrast algorithms are not based on perceptually uniform 110 110 lightness difference and that simple contrast does not consider the logarithmic 111 111 response characteristics of the human eye are bogus. 112 112113 -1 # APCA-1 113 ## APCA 114 114 115 115 The final contrast formula mentioned in the color.js documentation is APCA. If 116 116 you are interested in that I can refer you to my [detailed analysis][3]. 117 117118 -1 # Conclusion-1 118 ## Conclusion 119 119 120 120 The main takeaway is that simple contrast (and therefore WCAG 2.1) is not as 121 121 bad as some people make it sound. It *is* based on a difference of perceptual
diff --git a/_content/posts/2022-12-03-my-window-manager-setup/index.md b/_content/posts/2022-12-03-my-window-manager-setup/index.md
@@ -2,7 +2,7 @@ I am still on X11 because I haven't yet found the right wayland setup for me. 2 2 In this process, I had to take a good look at my current X11 setup. I was 3 3 surprised what I found. 4 45 -1 # My current setup-1 5 ## My current setup 6 6 7 7 I usually have all windows maximized, so only a single window is visible at a 8 8 time. (I also only have a single screen.) If I want to move a window across the @@ -43,7 +43,7 @@ browser `Ctrl-t` `Ctrl-w` `Ctrl-PageUp|PageDown` 43 43 terminal `Ctrl-Shift-t` `Ctrl-d` `Ctrl-PageUp|PageDown` 44 44 vim `:tabnew` `:wq` `Ctrl-Alt-PageUp|PageDown` 45 4546 -1 # How common is this?-1 46 ## How common is this? 47 47 48 48 I believe this is not too exotic. I pretty much stayed with the defaults and 49 49 the shortcuts like `Alt-Tab` or `Alt-F4` haven't really changed since I started @@ -61,7 +61,7 @@ issue with maximize-by-default is that dialogs should actuall not be maximized, 61 61 and window managers often have a hard time telling them apart. I am sure most 62 62 people who use tiling window managers can relate. 63 6364 -1 # Consistency-1 64 ## Consistency 65 65 66 66 What was shocking to me about this analysis was how inconsistent this all was. 67 67 Not just are there 5 different window managers, they all use different @@ -88,7 +88,7 @@ the browser, I can re-open the last recently closed tab. By forcing all these 88 88 levels into a single window manager I would loose those small but important 89 89 features. 90 9091 -1 # Keyboard usability-1 91 ## Keyboard usability 92 92 93 93 One downside I definitely see in my current approach is that it requires a lot 94 94 of different keys. `PageUp` and `PageDown` are not easy to reach on many @@ -98,7 +98,7 @@ two devices, so I constantly have to readjust. Also, the heavy use of modifier 98 98 keys makes it hard to use these shortcuts with a single hand, which might 99 99 become an accessibility concern. 100 100101 -1 # Why not use tiling?-1 101 ## Why not use tiling? 102 102 103 103 Displaying two windows side by side works on a stacking window managers, but it 104 104 also has its flaws: @@ -119,7 +119,7 @@ at all common there. 119 119 Tiling window managers might have benefits for my use case, but so far there 120 120 was no compelling reason to switch. 121 121122 -1 # Conclusion-1 122 ## Conclusion 123 123 124 124 I took a good look at my window management setup and found that is messy, but 125 125 sticks with defaults and works well.
diff --git a/_content/posts/2023-01-15-state-of-wayland/index.md b/_content/posts/2023-01-15-state-of-wayland/index.md
@@ -12,7 +12,7 @@ Please do not read this as a criticism of wayland as a whole. I understand that 12 12 my use case may not be super common. These are just my personal notes on how 13 13 well wayland works *for me*. 14 1415 -1 # The setup-1 15 ## The setup 16 16 17 17 My OS is Debian Bookworm, so all software is in the version that is included 18 18 with that OS. @@ -26,7 +26,7 @@ sway-bar with a custom status script. In addition there are several smaller 26 26 tools that are somewhat needed for a complete desktop: swayidle, swaylock, 27 27 grim, slurp, wl-clipboard, and xdg-desktop-portal-wlr. 28 2829 -1 # What improved?-1 29 ## What improved? 30 30 31 31 - Last time I had major issues with blurry fonts because I did not manage to 32 32 configure antialiasing correctly. This time around, `gsettings set @@ -60,7 +60,7 @@ grim, slurp, wl-clipboard, and xdg-desktop-portal-wlr. 60 60 than that the migration to gtk3 seems to be completed. So lets start porting 61 61 everything to gtk4 now! 62 6263 -1 # What needs more work?-1 63 ## What needs more work? 64 64 65 65 - GUI applications that run as root, such as gparted or synaptic, fail to 66 66 start. AFAIU, this is expected because of the stricter security model in @@ -115,7 +115,7 @@ grim, slurp, wl-clipboard, and xdg-desktop-portal-wlr. 115 115 already implemented in libinput and gtk4, but it [will not be backported to 116 116 gtk3](https://gitlab.gnome.org/GNOME/gtk/-/merge_requests/3454#note_1426517). 117 117118 -1 # Conclusion-1 118 ## Conclusion 119 119 120 120 So there you have it. 14 years after the initial wayland release we do have a 121 121 usable desktop experience. Unfortunately, there are still gaping holes.
diff --git a/_content/posts/2023-01-29-python-async-loops/index.md b/_content/posts/2023-01-29-python-async-loops/index.md
@@ -18,12 +18,12 @@ enjoy accompanying me on that journey. 18 18 If you are interested, all eight implementations are available on 19 19 [github](https://github.com/xi/python_async_loops). 20 2021 -1 # Setup-1 21 ## Setup 22 22 23 23 The following script outputs random numbers at random intervals: 24 24 25 25 ```bash26 -1 #!/bin/bash-1 26 ##!/bin/bash 27 27 while true; do 28 28 sleep $(($RANDOM % 5)) 29 29 echo $RANDOM @@ -85,7 +85,7 @@ def render(): 85 85 86 86 With these basics out of the way, we can start with the loop itself. 87 8788 -1 # Implementation 1: Blocking Loop-1 88 ## Implementation 1: Blocking Loop 89 89 90 90 Putting it all together we get our first implementation: The blocking loop: 91 91 @@ -143,7 +143,7 @@ In this version, `reader.read_line()` will block until data is available. So it 143 143 will first wait for data from `proc1`, then wait for data from `proc2`, then 144 144 render, repeat. This is not really async yet. 145 145146 -1 # Implementation 2: Busy Loop-1 146 ## Implementation 2: Busy Loop 147 147 148 148 ```python 149 149 import fcntl @@ -174,7 +174,7 @@ get stuck on a blocking read. However, this loop will just keep trying and 174 174 fully saturate the CPU. This is called a busy loop and obviously not what we 175 175 want. 176 176177 -1 # Implementation 3: Sleepy Loop-1 177 ## Implementation 3: Sleepy Loop 178 178 179 179 ```python 180 180 import time @@ -199,7 +199,7 @@ quickly we introduce a very noticeable delay of 1 second. And if data arrives 199 199 slowly we wake up much more often than would be needed. We can adjust the sleep 200 200 duration to the specific case, but it will never be perfect. 201 201202 -1 # Implementation 4: Select Loop-1 202 ## Implementation 4: Select Loop 203 203 204 204 ```python 205 205 import selectors @@ -233,7 +233,7 @@ So with this implementation we have our first *real* async loop, and from here 233 233 on out we will stick with selectors and only restructure the code surrounding 234 234 them. 235 235236 -1 # Implementation 5: Callback Loop-1 236 ## Implementation 5: Callback Loop 237 237 238 238 ```python 239 239 class Loop: @@ -307,7 +307,7 @@ You can register files with callbacks that will be executed whenever the file 307 307 is ready. There is also a much more sophisticated system for timeouts and 308 308 intervals, similar to what you might know from JavaScript. 309 309310 -1 # Aside: Everything is a File-1 310 ## Aside: Everything is a File 311 311 312 312 So far our loops can react to files and timeouts, but is that enough? My first 313 313 hunch is that in unix, "everything is a file", so this should get us @@ -355,12 +355,12 @@ pretty far. But let's take a closer look. 355 355 tells you that the underlying file descriptor is empty, but there is still 356 356 data available in the python buffer. 357 357358 -1 # Implementation 6: Generator Loop-1 358 ## Implementation 6: Generator Loop 359 359 360 360 We are getting closer to asyncio, but there is still a lot of conceptual ground 361 361 to cover. Before we get to async/await, we have to talk about generators. 362 362363 -1 ## Motivation-1 363 ### Motivation 364 364 365 365 As I said in the introduction, I personally really like the callback approach. 366 366 It is simple, just a selector and some callback functions. Compared to that I @@ -419,7 +419,7 @@ share state, and do cleanup. The callback approach tries to get by without 419 419 these benefits. The async/await approach instead tries to keep them by allowing 420 420 to pause the execution of functions. 421 421422 -1 ## The yield expression-1 422 ### The yield expression 423 423 424 424 The yield expression has been part of python since 425 425 [PEP 255](https://peps.python.org/pep-0255/) (2001) and got extended @@ -438,10 +438,10 @@ def foo(): 438 438 for x in foo(): 439 439 print(x) 440 440441 -1 # yielding 1442 -1 # 1443 -1 # yielding 2444 -1 # 2-1 441 ## yielding 1 -1 442 ## 1 -1 443 ## yielding 2 -1 444 ## 2 445 445 ``` 446 446 447 447 Another common use is to define context managers: @@ -552,7 +552,7 @@ For a more in-depth discussion of generators I can recommend the [introduction 552 552 to async/await by Brett 553 553 Cannon](https://snarky.ca/how-the-heck-does-async-await-work-in-python-3-5/). 554 554555 -1 ## The Loop-1 555 ### The Loop 556 556 557 557 ```python 558 558 import datetime @@ -716,7 +716,7 @@ pass through the messages by using `yield from`. 716 716 All this is mediated by `Task` objects which keep track of the conditions and 717 717 state of generators. 718 718719 -1 # Implementation 7: async/await Loop-1 719 ## Implementation 7: async/await Loop 720 720 721 721 From here it is a small step to async/await. Generators that are used for 722 722 asynchronous execution have already been called "coroutines" in PEP 342. [PEP @@ -892,7 +892,7 @@ async def amain(): 892 892 run(amain()) 893 893 ``` 894 894895 -1 # Implementation 8: asyncio-1 895 ## Implementation 8: asyncio 896 896 897 897 So which kinds of loop does asyncio use? After reading [PEP 898 898 3156](https://peps.python.org/pep-3156/) I would say: That's complicated. @@ -984,7 +984,7 @@ wrap loop callbacks. 984 984 985 985 This approach works reasonably well. But I also see some issues with it. 986 986987 -1 ## Limited support for files-1 987 ### Limited support for files 988 988 989 989 You may have noticed that I did not implement the full subprocess example this 990 990 time. This is because asyncio's coroutine layer doesn't really support files. @@ -994,7 +994,7 @@ callbacks are called every time data is available for reading. This disconnect 994 994 can probably be bridged somehow, but this post is already long enough and I 995 995 didn't want to go down yet another rabbit hole. 996 996997 -1 ## Futures are not a monad-1 997 ### Futures are not a monad 998 998 999 999 If you know some JavaScript you have probably come across Promises. Promises 1000 1000 are basically the JavaScript equivalent of Futures. However, they have a much @@ -1032,7 +1032,7 @@ future2.add_done_callback(print_done) 1032 1032 loop.run_until_complete(future2) 1033 1033 ``` 1034 10341035 -1 ## Naming Confusion-1 1035 ### Naming Confusion 1036 1036 1037 1037 So far we have "Coroutines", "Futures", and "Tasks". The asyncio documentation 1038 1038 also uses the term "Awaitables" for anything that implements `__await__()`, so @@ -1057,7 +1057,7 @@ final part could explain the low level plumbing for those people who want to 1057 1057 dive deep. This is the only part that needs to mention terms like "Awaitable", 1058 1058 "Task", or "Future". 1059 10591060 -1 # Conclusion-1 1060 ## Conclusion 1061 1061 1062 1062 These were eight different versions of asynchronous loops. I have certainly 1063 1063 learned something. A bit about async primitives on linux and a lot about
diff --git a/_content/posts/2023-02-26-flow-relative-layout/index.md b/_content/posts/2023-02-26-flow-relative-layout/index.md
@@ -19,7 +19,7 @@ current writing mode. Unfortunately it is still in the draft stage. Some parts 19 19 are already available in browsers, but some other parts are still [likely to 20 20 change](https://www.w3.org/TR/css-logical-1/#issue-3d880eb1). 21 2122 -1 # What works-1 22 ## What works 23 23 24 24 - `block-size` and `inline-size` 25 25 - `margin-block-start`, `margin-inline-end`, … @@ -33,7 +33,7 @@ change](https://www.w3.org/TR/css-logical-1/#issue-3d880eb1). 33 33 34 34 Source: <https://www.w3.org/TR/CSS/#CR-exceptions> and <https://caniuse.com/css-logical-props> 35 3536 -1 # What doesn't work-1 36 ## What doesn't work 37 37 38 38 - [`float`](https://caniuse.com/mdn-css_properties_float_flow_relative_values) 39 39 - [`overflow-block|inline`](https://caniuse.com/mdn-css_properties_overflow-block) @@ -49,7 +49,7 @@ with the layout but that do not adapt to the writing mode for a good reason: 49 49 - shadow offsets refer to physical directions 50 50 - … 51 5152 -1 # Recommendations-1 52 ## Recommendations 53 53 54 54 There is no real reason to use `width` or `margin-bottom` anymore. Just get 55 55 used to `inline-size` and `margin-block-end`.
diff --git a/_content/posts/2023-03-30-tui/index.md b/_content/posts/2023-03-30-tui/index.md
@@ -7,13 +7,13 @@ Because TUIs and CLIs are so different, switching between the two modes can be 7 7 challenging. If you are not careful, the terminal ends up in an inconsistent 8 8 state. But fear not! I will walk you through all the relevant steps. 9 910 -1 # Enter and exit TUI mode-1 10 ## Enter and exit TUI mode 11 11 12 12 Terminals do not have a dedicated TUI mode. Instead, there is a collection of 13 13 options we can combine to get what we want. I will first describe each aspect 14 14 and then provide a code example. 15 1516 -1 ## Disable line buffering-1 16 ### Disable line buffering 17 17 18 18 In CLI mode, input is buffered and sent to `stdin` when the user presses the 19 19 enter key. In TUI applications, we want to react to every key press @@ -30,7 +30,7 @@ The python standard library also provides bindings for termios as well as the 30 30 higher level `tty` module that allows us to easily disable line buffering using 31 31 the `cbreak()` function. 32 3233 -1 ## Restore screen content on exit-1 33 ### Restore screen content on exit 34 34 35 35 We do not want our TUI to mess up the terminal's scrollback buffer. So when we 36 36 exit, we want to restore the screen content as it was before we started. @@ -54,7 +54,7 @@ Switching to the alternate screen is `smcup` in terminfo and `\033[?1049h` in 54 54 xterm. Switching back to the original screen is `rmcup` in terminfo and 55 55 `\033[?1049l` in xterm. 56 5657 -1 ## Hide the cursor-1 57 ### Hide the cursor 58 58 59 59 When typing in the command line, the cursor show us where the next typed 60 60 character will be inserted. In many TUI applications, arbitrary regions of the @@ -65,7 +65,7 @@ This can again be done using escape codes. Hiding the cursor is `civis` in 65 65 terminfo and `\033[?25l` in xterm. Showing the cursor is `cnorm` in terminfo 66 66 and `\033[?25h` in xterm. 67 6768 -1 ## Putting it all together-1 68 ### Putting it all together 69 69 70 70 ```python 71 71 import sys @@ -92,14 +92,14 @@ run_mainloop() 92 92 exit_tui() 93 93 ``` 94 9495 -1 # Handling exceptions-1 95 ## Handling exceptions 96 96 97 97 The above code still has a major issue: When our mainloop raises an exception, 98 98 the process ends without exiting TUI mode, so we end up with broken terminal. 99 99 The fix in this case is simple though: Wrap the code in a `try … finally` block 100 100 so the cleanup code is run even if there are exceptions. 101 101102 -1 # Handling `ctrl-z`-1 102 ## Handling `ctrl-z` 103 103 104 104 You can stop any program in the terminal by pressing `ctrl-z`. That program 105 105 will simply not do anything until you type `fg`. When we stop our TUI @@ -148,7 +148,7 @@ Some things to note: 148 148 - The screen might have changed in the meantime, so it is best to do a fresh 149 149 render. 150 150151 -1 # Using context managers-1 151 ## Using context managers 152 152 153 153 If you are like me, the code examples above scream *context manager*. This 154 154 could look something like this: @@ -202,7 +202,7 @@ i.e. we must be able to enter and exit it multiple times. This is the case 202 202 here, but is not always guaranteed. For example, context managers that are 203 203 created using the `@contextmanager` decorator are not reusable. 204 204205 -1 # Get terminal size-1 205 ## Get terminal size 206 206 207 207 As the cherry on top, we should know how much space is available, e.g. to draw 208 208 a bar that spans the complete width of the terminal. Python provides us with a @@ -212,7 +212,7 @@ However, terminals can be resized. So you should also register a handler for 212 212 `SIGWINCH` that gets the latest size and re-renders your application whenever 213 213 the size changes. 214 214215 -1 # Conclusion-1 215 ## Conclusion 216 216 217 217 There are many libraries that take care of all of this for you. But sometimes 218 218 you run into issues anyway. You should now have all the tools to track down the
diff --git a/_content/posts/2023-09-28-mfa/index.md b/_content/posts/2023-09-28-mfa/index.md
@@ -10,7 +10,7 @@ old "passwords are bad and MFA is better" dance. I will still do the 10 10 comparison, but only skim the boring parts and quickly move on to the 11 11 (hopefully) more interesting details. 12 1213 -1 # Passwords-1 13 ## Passwords 14 14 15 15 Passwords have been used for authentication for a long time, even though they have many issues: 16 16 @@ -23,13 +23,13 @@ Passwords have been used for authentication for a long time, even though they ha 23 23 24 24 In short: Passwords are in too many places. 25 2526 -1 # Two Factor Authentication (2FA)-1 26 ## Two Factor Authentication (2FA) 27 27 28 28 Now, replacing passwords all at once didn't seem realistic. Also, all the 29 29 alternatives we had available came with issues of their own. So instead of 30 30 replacing passwords, we started adding other authentication mechanisms on top. 31 3132 -1 # One-Time Passwords-1 32 ## One-Time Passwords 33 33 34 34 A very common approach to two factor authentication is to send one-time 35 35 passwords over a side channel such as @@ -45,7 +45,7 @@ phishing is easy: Attackers can just call users and say "I need you to verify 45 45 your identity by telling me the current code" and they are in. Also, the 46 46 symmetric secrets cannot be hashed on the server. 47 4748 -1 # Something you own-1 48 ## Something you own 49 49 50 50 Many people say that the two factors in "two factor authentication" are 51 51 knowledge and possession. Sometimes they also add biometrics ("something you @@ -70,7 +70,7 @@ are not perfect. I fear we might get into trouble if our crypto protocols start 70 70 to depend on TPMs. Instead, we should think of them as obstacles. They make 71 71 attacks harder, but don't give any guarantees. 72 7273 -1 # Multi Factor Authentication (MFA)-1 73 ## Multi Factor Authentication (MFA) 74 74 75 75 My impression is that the term "2FA" is slowly being displaced by "MFA". The 76 76 "multi" is not so much about "two or more factors", but more about a move to a @@ -83,7 +83,7 @@ This also means that we are in a time of change, kinda post-password. We 83 83 already know that passwords are a thing of the past, but we have not yet 84 84 settled on a new approach. 85 8586 -1 # Single Sign-On (SSO)-1 86 ## Single Sign-On (SSO) 87 87 88 88 Let's shift gears for a moment and think about SSO. 89 89 @@ -100,7 +100,7 @@ website (*relying party*), but the actual authentication is done by an 100 100 mechanisms (long random passwords, challenge-response, asymmetric 101 101 cryptography, …). 102 102103 -1 # Local Identity Providers-1 103 ## Local Identity Providers 104 104 105 105 If the identity provider is running on your devce, the authentication doesn't 106 106 need to involve the network, which significantly reduces the attack surface. @@ -114,7 +114,7 @@ Your SSH keys also act as local identity providers. Compared to password 114 114 managers, they use a much better protocol to communicate with the relying 115 115 party. 116 116117 -1 # WebAuthn / FIDO2-1 117 ## WebAuthn / FIDO2 118 118 119 119 [WebAuthn](https://www.w3.org/TR/webauthn/) is a new JavaScript interface. Even 120 120 though it is often talked about in the context of MFA, I believe it is more @@ -152,7 +152,7 @@ cryptography to communicate between the authenticator and the relying party and 152 152 that it has some other benefits (e.g. use of TPMs) that even put it ahead of 153 153 SSH in terms of security. 154 154155 -1 # Client-side MFA-1 155 ## Client-side MFA 156 156 157 157 So far we have discussed that WebAuthn is a form of local SSO. It can be used 158 158 as a building block for MFA either by using it as one of several authenticaion @@ -181,7 +181,7 @@ Apart from user verification, authenticators can also check user presence. 181 181 However, relying parties have no control over that and it is always required. 182 182 That means that authentication without user interaction is not possible. 183 183184 -1 # Moving beyond passwords-1 184 ## Moving beyond passwords 185 185 186 186 As explained in the beginning, the goal is to get rid of passwords completely. 187 187 The buzzword a year ago was "passwordless", the current iteration is "passkey". @@ -212,7 +212,7 @@ the same relying party becomes a bit more complicated. Either I have different 212 212 authenticators, or the authenticator allows me to pick one, or the relying 213 213 party receives all keys and asks me which one to use. 214 214215 -1 # Recovery-1 215 ## Recovery 216 216 217 217 Before ending this post, we also have to look into some potential issue. The 218 218 first one is recovery. @@ -236,7 +236,7 @@ the same side channel or recovery keys are stored in the same password manager 236 236 as the password. If that is the case, you are not doing MFA and not gaining any 237 237 security. So make sure that your recovery options are solid! 238 238239 -1 # Account enumeration attacks-1 239 ## Account enumeration attacks 240 240 241 241 When we are thinking about password-less approaches we also have to consider 242 242 some attacks that have been solved for passwords a long time ago. @@ -252,7 +252,7 @@ enumeration in that case? The 252 252 creating a fake key in that case, but that sounds like there could be all kinds 253 253 of footguns. 254 254255 -1 # Conclusion-1 255 ## Conclusion 256 256 257 257 After reading and hearing a lot about MFA, I had to structure all that 258 258 information. I believe the idea to think of WebAuthn as local SSO can be
diff --git a/_content/posts/2023-10-13-logging/index.md b/_content/posts/2023-10-13-logging/index.md
@@ -5,7 +5,7 @@ than would be needed for such a simple task. But this library also offers a lot 5 5 of flexibility for advanced users as well as shortcuts for simple use cases. So 6 6 let's take a look. 7 78 -1 # Creating log records-1 8 ## Creating log records 9 9 10 10 Typically you would create log records like this: 11 11 @@ -35,7 +35,7 @@ instead, e.g. `logging.error()`. These will use the root logger. If handlers 35 35 have not yet been configured (see next section), these shortcuts will also 36 36 make sure that log records are written to stderr. 37 3738 -1 # Configuring handlers-1 38 ## Configuring handlers 39 39 40 40 Log records are sent to stderr by default. In production settings you usually 41 41 want something more involved, so you have to configure handlers. @@ -128,7 +128,7 @@ can receive configuration in JSON format. I am not sure why you would want to 128 128 do that and I find it rather strange that this is included in the standard 129 129 library. 130 130131 -1 # Attaching handlers to specific loggers-1 131 ## Attaching handlers to specific loggers 132 132 133 133 In most cases you will want to attach your handlers to the root logger and then 134 134 filter records there. But you can also attach handlers to different loggers. @@ -162,7 +162,7 @@ logging.config.dictConfig({ 162 162 This way, log records from the "audit" namespace will be logged to a file 163 163 instead of stderr. 164 164165 -1 # Recommendations-1 165 ## Recommendations 166 166 167 167 `logging` has lots of flexibility, so much so that it is sometimes hard to know 168 168 how to do simple things. So here are some recommendations: @@ -178,7 +178,7 @@ how to do simple things. So here are some recommendations: 178 178 of `WARNING`. To avoid confusion, I strongly recommend to reset that to 179 179 `NOTSET`. 180 180181 -1 # Wishful thinking-1 181 ## Wishful thinking 182 182 183 183 With that said, there are some things in `logging` that I wish were different: 184 184 @@ -198,7 +198,7 @@ With that said, there are some things in `logging` that I wish were different: 198 198 It adds a lot of complexity without a tangible benefit because you could 199 199 achieve the same thing with filters. 200 200201 -1 # Conclusion-1 201 ## Conclusion 202 202 203 203 The right logging configuration is crucial. If your logs get lost you cannot 204 204 debug issues. And if you do not get notified about errors, you might not even
diff --git a/_content/posts/2023-10-29-jmap-calendars/index.md b/_content/posts/2023-10-29-jmap-calendars/index.md
@@ -1,4 +1,4 @@1 -1 # The old thing: CalDAV-1 1 ## The old thing: CalDAV 2 2 3 3 The predominant protocol used for syncing calendars is CalDAV 4 4 ([RFC 4791](https://www.rfc-editor.org/rfc/rfc4791)). CalDAV is a weird @@ -95,7 +95,7 @@ Frustrated with all those issues, I started thinking about a new protocol that 95 95 could replace CalDAV. After a bit of research, I found that someone had already 96 96 done that work. 97 9798 -1 # The new thing: JMAP Calendars-1 98 ## The new thing: JMAP Calendars 99 99 100 100 [JMAP Calendars](https://jmap.io/spec-calendars.html) is a draft spec to send 101 101 events in JSCalendar format