Replies: 7 comments 2 replies
-
This test succeeds on our MUSL CI builds and in our 32-bit CI builds. The
difference seems to be this line:
* Cannot rewind mime/post data
which doesn't happen when the test succeeds. The test program stops at that
point so the rest of the transfer lines aren't sent resulting in a truncated
file and that large diff. I don't know offhand why that would occur in your
case and not in mine; whether it succeeds should be a factor of how the test is
set up and not the platform.
|
Beta Was this translation helpful? Give feedback.
-
The test also passes on Alpine Linux 32 bit x86 |
Beta Was this translation helpful? Give feedback.
-
I created a logfile using
|
Beta Was this translation helpful? Give feedback.
-
I bisected it. The commit that introduces this is 74b87a8 |
Beta Was this translation helpful? Give feedback.
-
That matches what was found in #11900 |
Beta Was this translation helpful? Give feedback.
-
I have gdb session running here and it appears that res = part->seekfunc(part->arg, (curl_off_t) 0, SEEK_SET); where seekfunc is fseek.
It appears that 22 is
So
So it is fd 0, which means In other words it is Not sure why this happens on arm32 only though. Maybe stdin is a pipe and the FILE* buffer is smaller on arm32? |
Beta Was this translation helpful? Give feedback.
-
Using a breakpoint on
This happens because seekfunc takes a curl_off_t but fseek takes a To fix this we can either use a wrapper function, or use Thanks alot to the nice people in Libera IRC #musl for help finding this out. |
Beta Was this translation helpful? Give feedback.
{{title}}
-
Hello, while testing curl 8.3.0 in Alpine Linux's Gitlab CI, i ran into issues with test 650 on 32-bit ARM.
Alpine Linus uses musl libc, and here is curl's info:
This is the output from test 650. I would appreciate it if someone could tell me if this looks safe to ignore, or if it could be indicative of a more serious bug.
Beta Was this translation helpful? Give feedback.
All reactions