aboutsummaryrefslogtreecommitdiff
path: root/sysdeps/generic/fix-fp-int-convert-overflow.h
diff options
context:
space:
mode:
authorZack Weinberg <zack@owlfolio.org>2023-09-21 14:45:59 -0400
committerZack Weinberg <zack@owlfolio.org>2023-09-21 16:05:43 -0400
commitf59011763cae16dc89afc87d9218c5937bf7686b (patch)
tree107003cb881d25ea615c25cb15664b889a1ed5da /sysdeps/generic/fix-fp-int-convert-overflow.h
parent5d00c201b9a2da768a79ea8d5311f257871c0b43 (diff)
downloadglibc-f59011763cae16dc89afc87d9218c5937bf7686b.tar
glibc-f59011763cae16dc89afc87d9218c5937bf7686b.tar.gz
glibc-f59011763cae16dc89afc87d9218c5937bf7686b.tar.bz2
glibc-f59011763cae16dc89afc87d9218c5937bf7686b.zip
Import Solar Designer's public domain MD5 for use by localedef.
Locale archives contain embedded MD5 hashes for integrity protection. glibc's locale-reading code does not check these, but localedef does generate them. It was reusing crypt/md5.c for the implementation. Rather than moving that file over to locale/, import Alexander Peslyak (aka Solar Designer)'s public domain MD5 implementation, which is simpler, and in particular, completely agnostic to endianness. The API uses different names, because Peslyak wanted to be API-compatible with openssl, but is otherwise equivalent. glibc's *tests* of the MD5 core (crypt/md5test.c and crypt/md5test-giant.c) are transferred to the locale directory, and the new implementation is verified to pass both. (The "giant" test takes 90 seconds to run on a 2018-era x86; it was in xtests in crypt and it remains in xtests after this patch.) I converted both of them to the new test driver while I was in there. crypt/md5c-test.c is a test of MD5 *password hashing*, not of the MD5 core, so it is not moved. This patch was compile-tested with both --enable-crypt and the default --disable-crypt.
Diffstat (limited to 'sysdeps/generic/fix-fp-int-convert-overflow.h')
0 files changed, 0 insertions, 0 deletions