On Mon, 16 Jul 2018, Dan Fandrich wrote:
This test has been failing on every single autobuild where it's run for a long time, probably since commit a115c6bb which switched libcurl to using the system fnmatch back in May.
Yeah, clearly...
I find it hard to understand why no-one has noticed for a month and a half since the Travis builds et. al. must have been succeeding during that time. Maybe it's a sign that the Travis builds are too monocultural and that the autobuilds really are adding some value (if anybody were to notice them).
I think I'm one of the worst offenders of mostly considering the travis builds and when they're green I'm good and move on.
In this specific case, I was clearly a bit naive in thinking that manually verifying three different combinations (private, Linux glibc and mac) would be enough, since it seems uclibc fnmatch() returns different results! And since I don't monitor the autobuilds closely enough, it avoided my attention.
Additionally, the fnmatch() functionality is only used by ftp wildcard matching which isn't used very much so I presume no user has found any problems with this change either...
I would love to see people help me out and file issues for failed autobuilds. At least such failures that used to work fine before.
-- / daniel.haxx.se ------------------------------------------------------------------- Unsubscribe: https://cool.haxx.se/list/listinfo/curl-library Etiquette: https://curl.haxx.se/mail/etiquette.html
