IRC Logs for #crux-devel Sunday, 2019-03-17

ryuohm. this error handler seems like a decent design.00:32
ryuoreport the error to the application and return gracefully to the application.00:33
ryuowow... this is dangerous...00:41
ryuoi may be mistaken but it appears pkgutils has no way to recover from a failure to extract the entire package to disk...00:41
ryuoi think i will correct this.00:42
ryuoi have an idea for how to address it.00:42
ryuoi'll extract to a temporary file name in each directory...00:42
ryuoif all package files extract successfully...00:43
ryuoi can do a pass to rename them to their real file name, unlinking the original.00:43
ryuoif any fail to write to disk, i can remove them to restore the root to its original state.00:43
ryuoat the very least pkgadd should *try* to recover from a failure to install/upgrade a package.00:45
ryuoit's next to impossible to accurately predict if the disk has enough space, so the next best option is to extract it to secondary files so a rollback is possible.00:47
ryuothe database doesn't just encompass the text metadata...00:47
ryuoit also encompasses the package contents.00:47
ryuohas to be same directory to avoid the dreaded EXDEV error that can occur if the rename operation crosses directories...00:48
ryuorename operations only work within the same filesystem; if directories are on different mount points it will fail.00:48
ryuoand only same directory is guaranteed to be the same filesystem or mount point.00:49
ryuoscrew it. i'm better off deferring the processing of system call errors to libc.01:05
ryuoi'll just tag them with a special bit flag.01:05
ryuook. that's better.01:06
ryuoif nothing else i should try to make this library as fault tolerant as possible... even for things that pkgutils doesn't already try to handle.01:16
ryuogiven that the package database must remain ordered...01:21
ryuoi think i will use the libc binary tree functions.01:21
ryuosince i need to be more flexible with possible modifications.01:22
Romsterjaeger, jue is core/libusb-compat needed anymore?06:56
Romstercore/openssl The INSTALL_PREFIX Makefile variable has been renamed to DESTDIR. work/src/openssl-1.1.1b/CHANGES07:26
Romstermake INSTALL_PREFIX=$PKG LIBDIR=lib MANDIR=/usr/share/man MANSUFFIX=ssl DESTDIR=$PKG install07:27
Romsterredundant INSTALL_PREFIX=$PKG07:27
Romster;a=blob;f=openssl-32/Pkgfile;h=9e64c2352aa203a3d91a922b5087692e68e4296f;hb=361c84f8f0d25742f49ac14dfe1bda6f16a38737 openssl-32/Pkgfile you don't even need LIBDIR=lib MANDIR=/usr/share/man07:56
Romstermake MANSUFFIX=ssl DESTDIR=$PKG install07:56
Romsterexport MAKEFLAGS="$MAKEFLAGS -j1" is also not needed anymore, just on make -j1 MANSUFFIX=ssl DESTDIR=$PKG install07:58
Romsteralso see i use /Configure and not ./config07:58
jueRomster: thx, will look11:27
*** onodera has joined #crux-devel12:31
*** onodera has quit IRC12:40
jueRomster: I've pushed a improved openssl to 3.5, see commit-message for comments12:47
jueRomster: guess it's save to remove libusb-compat, I've fixed deps of libgphoto2 already, what remains is opt/bluez, which seems to be an ancient version12:49
ryuono wonder it's so old. the maintainer got the bluez.12:50
jueany objections to remove core/libusb-compat from 3.5?12:50
ryuowhat would still use it? i suspect only unpatchable blobs at this point.12:51
juewell, we can move it to opt, just in case someone needs it12:53
ryuough. glibc, why?13:19
ryuoit provides two versions of strerror_r and only the POSIX one is what i want...13:20
ryuoa possible pitfall so i added some code that'll produce a compiler warning if the GNU version is somehow chosen.13:21
Romsterthat looks cleaner jue happy with that.13:56
*** onodera has joined #crux-devel13:59
jaegerjue: none from me14:29
ryuoi'm expanding the file info from the db to include other info...14:43
ryuoit seems pkginfo doesn't try to check for file existence as it loads the database...14:43
ryuoto do so would likely create problems, but i could add that as an integrity check mode later14:43
ryuoto do so now*14:43
ryuomodes to report files that are missing from disk or aren't in the database yet exist14:44
ryuough. i need more dynamic arrays than i realized.16:41
ryuoi'm going to make a generic setup for them.16:41
ryuomuch better. i took inspiration from how qsort, bsearch, lsearch work.17:04
*** onodera has quit IRC21:56
frinnstseems my freetype broke cairo, can someone else confirm=22:41
frinnst/bin/sed: can't read /usr/lib/ No such file or directory fun times. lets push glib/meson to 3.4 and watch the world burn :D22:42
pedjagood thing I have my own freetype port, then :)22:54

Generated by 2.14.0 by Marius Gedminas - find it at!