Zì

Windows argument passing

The thing about command-line on Windows is that it's the wild west. Programs do not actually receive an argv[] array like you have been conditioned by C to believe; instead they get a "cmdline" string to do whatever the hell they want with it.

Still, in that sea of insanity, there is one constant. Regular programmers are lazy. They just use whatever the runtime library does with the command line. In these cases you get a somewhat predictable behavior -- "somewhat", because even amount Microsoft runtimes the behavior differs -- and you can write some sort of an escaping function.

Beyond the runtime, cmd.exe also has its own understanding of how quoting works. Most disgustingly, it understands where double-quoted parts start and end differently from msvcrt. I've got a trick for that.

Node

The node PR, at https://github.com/nodejs/node/pull/29576, is focused on quoting. It's got proper quoting for bash, pwsh, and cmd + msvcrt. It's got very strong test cases (throw them at other people's things: they might break!) and good documentation. What it didn't have is communication and rizz. You're advised to skip to the "files changed" tab.

The cmd "trick" is explained at https://github.com/Artoria2e5/node/blob/c0a6aff35cd3ff6d3bd0e3687776158e97466c93/lib/child_process.js#L850. The gist is that msvcrt and .NET treats two double-quotes as if it's an escaped double-quote. It's undocumented, sure, but Microsoft's own code relies on it.

doc/api/child_process.md has a good chunk of text explaining how Windows cmdline works and who deviates from the norm. And that leads to our next suspect: Cygwin.

Cygwin

Cygwin is the way to run POSIX things on Windows these days. Sure there's MSYS2, but guess what? It's a Cygwin fork. If you are using the official Windows build of git, you're using MSYS2, which in turn means you're using Cygwin code.

Cygwin is not msvcrt, obviously. But what's less obvious is that it doesn't process cmdline the same way. That bites you HARD when you want to run a Cygwin program on a file with weird characters.

The patchset at https://github.com/mirror/newlib-cygwin/pull/5, formerly a botched e-mail exchange ending with me being too tired to split the patchset (come on, it's almost impossible to have intermediate stages that build, let alone work!), was my answer to the problem. It rewrites the cmdline parser to be msvcrt-compatible, while still keeping the Unix-style globbing and @-expansion features.

It's fucking awesome is what it was. It's written in C++ with in-place processing. What else can you ask for.

Oh, in case you want to know what my current answer to the problem is: roll into a ball and cry.

Openstep plistlib

So if you use an Apple product and do things that Apple don't really like you to do, you have probably came across a "plist". Nowadays plists are either an opaque binary file or a very wordy XML. Back in the good old days plists were not like that at all: they read more like JSON, except they only supported strings and dicts.

Python's plistlib does not care about the old style. It should, because it's fucking beautiful. It's also as functional as the new stuff is, as long as you use the GNUStep extensions.

I wrote a Python package to do just that (or did I? does it work yet?). https://github.com/Artoria2e5/text-plistlib As far as I can recall it works on a round-trip smoke test, but some extension decisions needed to be hammered out. It uses tatsu though, so there's no chance of getting into the plistlib part of the standard library.

I think I've also got some other things related to this plist format open somewhere. I wanted to write a parser for opencore to cut down on this XML misery, but COME ON IT'S C. I can't even keep track of a plain recursive descent parser in Python (why else am I using a parser generator?), to do it in C? No way. I'm also too lazy to learn yacc, thank you very much.

Other things

something about https://marc.info/?l=openssh-unix-dev&m=168509072920594&w=2

Dream the DRM receiver/broadcaster has bad Opus options. Dev said I should open a branch. I think I can't, because I can't even get the official binary to run. I also lost my slightly illegal SDR kit.