Hi Bruno,
On Sun, 11 Oct 2020, Bruno Haible wrote: > Tim Rice wrote: > > ./gnulib-tool --create-testdir --dir=/var/tmp --with-tests \ > > --single-configure --avoid=havelib-tests fseterr freadable fwritable \ > > fbufmode freading fwriting freadptr freadseek freadahead fpurge fseeko > > ftello fpending fflush > > Yes, this is the way to test these modules. > > > and the attached patch, I am down to 2 failures in the stdio tests. > > Hmm, there are three things I find suspicious: > > 1) __freadable is defined in a way that is inconsistent with Gnulib: > #define __freadable(p) (__freading(p) && __fpending(p)) > [in Gnulib, a stream that is not reading nor writing is considered > to be readable] > but you don't ensure that freadable doesn't get defined other than > __freadable. > > 2) Likewise for __fwritable, which is defined in a way inconsistent with > Gnulib: > #define __fwritable(p) (__fwriting(p) && __fpending(p) < __fbufsiz(p)) Indeed. And after the post with the defines I found tests for for __freadable and __fwritable were failing. It looks like we just need to check for _IOREAD, _IORW, _IOWRT flags. Sorry I didn't update the post with the defines. This is what I'm working with now which does pass __freadable and __fwritable tests. #define __freadable(p) ((void)sizeof(__filbuf(p)), ((FILE *)(p))->__flag & (_IOREAD | _IORW)) #define __fwritable(p) ((void)sizeof(__filbuf(p)), ((FILE *)(p))->__flag & (_IOWRT | _IORW)) > 3) In freadahead.c you add: > return __fpending(fp_); > But __fpending looks at the pending output, whereas freadahead shall look > at the available input. Should be different things, no? I had that change for so long I had to research why it was done that way. Looking back at a 2009 an e-mail from one of the SCO engineers, I see "... using the __fpending(), which in our libc.so.1 will work for both input or output files." > > Bruno > -- Tim Rice Multitalents t...@multitalents.net
