Randall S. Becker
Randall S. Becker
After the call time mktime() at 632 (ref above), the `tm_isdst` value is set to 1. It is 0 before. mktime is non-reentrant and can modify `struct tm`. Is there...
> > Is there an assumption that `tm_isdst` does not change? Might have to save it off first. > > No, that is not the problem. This is IMO actually...
Report from our CI system. Still broken as of e5f831a065df1d6e4640ef389f8594a5f10c9c8e. This defect still exists but no longer hangs the test suite.
This is still broken as of commit 8547cd6790
This appears to have been fixed by an unrelated change.
If I can comment on this, using `fopen()` instead of `stat()` has non-portable side-effects. `fopen()` modifies attributes in the file system (`st_atime`) and is a heavy-weight operation on some platforms....
@vszakats Thank you
Question on this case: Is there a mixture of static and dynamic linkage for OpenSSL? We have seen issues in OpenSSL when that is done. `atexit()` semantics with DLLs can...
> > > Because that second and thus incorrect call to OpenSSL_cleanup is from calling say curl_multi_cleanup. > > > > > > I don't see that in your stack...
> > I don't see that it is, `OpenSSL_cleanup` is being called first at exit and THEN later on during `curl_multi_cleanup` we land on a path that takes us through...