> ???? The above line strikes me as a typical fence-post error on the
> part of the application author.
No. Some Solaris, for example, use a char (not even unsigned char!) to
store the file descriptor in their FILE structure (E.g., what K&T (1st
ed.) shows as "int _fd" on page 165.
> Can it be shown that there are bugs for normal use?
Yes. I have a big multi-threaded app (running on a big multi-cpu
processor, presumably) that opens several hundred connections, each to a
unique client, and each with its own config file. It then calls
CONF_load, which calls BIO_new_file, which will fail once fopen fails --
at around the 500'th client, depending on timing.
I am skeptical, but can believe that fopen/et al are more portable than
open/read et al. Which is why I suggested making it a compile-time flag
in bss_file.c
> Also, have you looked at bio/bss_fd.c?
Sure, but the issue is what OpenSSL uses interally, as I hope my example
showed.
/r$
______________________________________________________________________
OpenSSL Project http://www.openssl.org
Development Mailing List [EMAIL PROTECTED]
Automated List Manager [EMAIL PROTECTED]
