[-- Attachment #1: Type: text/plain, Size: 699 bytes --] Hi, I'm not subscribed, please keep me CC'd. In this corner case, errno is unset despite fputs giving an error. I'm on Debian testing using musl-gcc 1.2.1. #define _POSIX_C_SOURCE 200809L #include <assert.h> #include <errno.h> #include <stdio.h> int main(void) { FILE *f = fopen("/", "r"); if(f == NULL) { perror("fopen()"); } if(fputs("Hello world\n", f) == EOF) { assert(errno); perror("fputs()"); } } With glibc this prints EBADF for fputs. The wiki page about writing tests [1], which I thought this would be a good candidate for, has the 404 link http://nsz.repo.hu/git/?p=libc-test Please let me know if I can help debugging. [1] https://wiki.musl-libc.org/writing-tests.html [-- Attachment #2: This is a digitally signed message part. --] [-- Type: application/pgp-signature, Size: 228 bytes --]
On Thu, Oct 08, 2020 at 10:55:47AM -0400, John Scott wrote:
> Hi,
>
> I'm not subscribed, please keep me CC'd.
>
> In this corner case, errno is unset despite fputs giving an error.
> I'm on Debian testing using musl-gcc 1.2.1.
>
> #define _POSIX_C_SOURCE 200809L
> #include <assert.h>
> #include <errno.h>
> #include <stdio.h>
> int main(void) {
> FILE *f = fopen("/", "r");
> if(f == NULL) {
> perror("fopen()");
> }
> if(fputs("Hello world\n", f) == EOF) {
> assert(errno);
> perror("fputs()");
> }
> }
>
> With glibc this prints EBADF for fputs.
>
> The wiki page about writing tests [1], which I thought this would be
> a good candidate for, has the 404 link
> http://nsz.repo.hu/git/?p=libc-test
>
> Please let me know if I can help debugging.
>
> [1] https://wiki.musl-libc.org/writing-tests.html
The test is invoking undefined behavior by calling an output function
on a stream not opened for write or update mode. If you instead opened
"/" with O_RDONLY but then passed it to fdopen with mode "r+" or
similar, the POSIX-specified "shall fail" for "The file descriptor
underlying stream is not a valid file descriptor open for writing"
would apply and EBADF would be set (assuming fdopen didn't take the
liberty to fail here, which POSIX allows but does not require and musl
does not do since it would make fdopen more expensive).
Rich
On Thu, Oct 08, 2020 at 10:55:47AM -0400, John Scott wrote: > Hi, > > I'm not subscribed, please keep me CC'd. > Here's hoping I chose the correct option this time. I went with "group reply" but directed the message to you, which should CC the list. > In this corner case, errno is unset despite fputs giving an error. I'm on Debian testing using musl-gcc 1.2.1. > > #define _POSIX_C_SOURCE 200809L > #include <assert.h> > #include <errno.h> > #include <stdio.h> > int main(void) { > FILE *f = fopen("/", "r"); > if(f == NULL) { > perror("fopen()"); > } > if(fputs("Hello world\n", f) == EOF) { > assert(errno); > perror("fputs()"); > } > } > > With glibc this prints EBADF for fputs. > Interesting. The error occurs because f is only open for reading, so __towrite() delivers an error without setting errno. Though my POSIX manpage does not say which error to return for this. EBADF speaks about the FD underlying the stream, and in this case that one happens to be the case, but the error occurs solely because F_NOWR is set in the file. The FD does not come into play. On the other hand, using fdopen() such that the mode given to open() and the string given to fdopen() are discordant is undefined behavior. So we might as well assume that F_NOWR means O_RDONLY. Besides, no other documented error on that page really fits this case. I suggest adding error codes to the error paths in __towrite() and __toread() (which has the same problem). > The wiki page about writing tests [1], which I thought this would be a good candidate for, has the 404 link http://nsz.repo.hu/git/?p=libc-test It would appear Szabolcs' repo is broken. I cannot access .../git/, but accessing .../ yields a landing page that references the non-working pages given here. I managed to find a mirror here: https://repo.or.cz/libc-test.git But that one also reference Szabolcs' repo as the original. Ciao, Markus
* Markus Wichmann <nullplan@gmx.net> [2020-10-08 18:23:34 +0200]: > On Thu, Oct 08, 2020 at 10:55:47AM -0400, John Scott wrote: > > > The wiki page about writing tests [1], which I thought this would be a good candidate for, has the 404 link http://nsz.repo.hu/git/?p=libc-test > > It would appear Szabolcs' repo is broken. I cannot access .../git/, but > accessing .../ yields a landing page that references the non-working > pages given here. I managed to find a mirror here: sorry about that, i fixed it now. (that website went through a migration that was left incomplete, now fixed) > > https://repo.or.cz/libc-test.git > > But that one also reference Szabolcs' repo as the original. git clone worked, but the gitweb browser did not.