On 08/05/2026 05:30, Paul Eggert wrote:
On 2026-05-06 12:09, Paul Eggert wrote:
It appears that test-regex.c is written
wrongly; its use of re.fastmap is undocumented. Possibly this means my
regex fix just lucked out and is not really a fix. I'll look into it.
After looking into it, it seems that this was entirely the fault of the
test program. I installed the attached patches into Gnulib to fix the
test issues I found.
I found some other annoyances with GCC 16's warnings spitting out both
false positives and good advice, notably that unless you need a constant
expression the compound literal ((type) {expr}) is often better than the
cast ((type) (expr)), as the cast has too much power.
I fixed all the GCC 16 warning issues I found in the Gnulib files copied
to coreutils/lib, and will send out emails to bug-gnulib shortly. Still
haven't gotten around to looking at the files copied to
coreutils/gnulib-tests, which are lower priority.
FYI test-regex-el is failing with latest rawhide (glibc) like:
test-regex: fastmap mismatch: pattern = \xd3, string = \xf2, with = -1,
without = 0
test-regex: fastmap mismatch: pattern = \xf2, string = \xf2, with = -1,
without = 0
test-regex: fastmap mismatch: pattern = \xf3, string = \xf2, with = -1,
without = 0
This seems to be due to the gnulib replacement now _not_ being used,
while there looks to have been no progress in glibc on the issue:
https://sourceware.org/bugzilla/show_bug.cgi?id=20381
Do we need an explicit check for the issue to enable the replacement?
cheers,
Padraig
