There's a 10+ years old libclang issue that I'm trying to figure out: "libclang does not get the header search correct"

I'm still hitting the issue on multiple distros. Debian carries a local fix for it.

What options should I pass to libclang to work around it? The key is automation; I need to be able to figure out the options programmatically, without a build system. (I'm using this from Python bindings.)

@nathanchance any ideas?

https://github.com/llvm/llvm-project/issues/18150

#libclang #clang #llvm

This is now kind of a dev microblog concerning #Xmoji. I'm kind of stalled, now that version 0.7 seems reasonably stable and portable (I see there's a #nix pkg, unfortunately outdated, and a #NetBSD #pkgsrc port, I will deliver #FreeBSD *soon*).

It misses a few convenience features my previous #qXmoji had: save/restore the window size, optionally enforce a single instance, offer a #tray icon. I'll add all of that, seems straight-forward, for the tray icon I'll only implement the old #X11 spec based on #Xembed and if some desktop environment insists on only supporting the newer standard based on #dbus, well, screw that. Too much complexity, sorry.

The real issue is #localization (#l10n), specifically "just" translations. I still have no good concept for that. With #Qt, it was a no-brainer to also use Qt's mechanism. Without a toolkit, obvious choices would be either #POSIX message catalogs, or #GNU #gettext. The latter is much more convenient, but pulls in extra deps (with #GPL/#LGPL foo). Both have in common that they only operate on char* ... 8bit encodings. I have many of my texts stored as char32_t (#Unicode UCS-4 or #UTF32, difference doesn't matter much here). I could redesign that to base everything on #UTF8, but I'm a bit reluctant ... why add more runtime conversions?

I seriously think about coming up with my own tooling. But then, how far should I jump? Should I really try to parse my own source (using LLVM's #libclang for example)? Or should I hardcode tables with identifiers for all translatable texts?

I'll sleep on that a few more nights I guess....

#X11 #emoji #keyboard

Atíla Neves ponders about
#UnitTesting. The #D programming language is too closely coupled to #libclang, and Neves is working on separating it. In the process Neves found the unit tests to be #integrationTests rather than individual tests. Neves held many tech talks telling us to avoid that. So Neves feels ashamed, and will be correcting the tests.

https://atilanevesoncode.wordpress.com/2018/11/20/unit-testing-do-as-i-say-dont-do-as-i-do/