Re: [musl] Broken freopen() does not reset fwide()

[Rich Felker <dalias@libc.org>]

On Wed, Aug 17, 2022 at 11:11:31AM -0400, Rich Felker wrote:
> On Tue, Aug 16, 2022 at 06:33:35PM +0000, Anonymousemail wrote:
> >    Powered by Anonymousemail → [1]Join Us!
> > 
> >    I'm using musl based distribution.
> >    The official example from https://en.cppreference.com/w/c/io/fwide is
> >    malfunctioning, musl does not reset the fwide(), returns -1 (meaning BYTE
> >    oriented).
> > 
> >    Output from example on musl:
> >    1) A newly opened stream has no orientation.
> >        no orientation
> >    2) Establish byte orientation.
> >        narrow orientation
> >        narrow character read '#'
> >        wide character read 'i'
> >    3) Only freopen() can reset stream orientation.
> >    4) A reopened stream has no orientation.
> >        narrow orientation  <- problem detected here, should say "no
> >    orientation"
> >    5) Establish wide orientation.
> >        narrow orientation
> >        narrow character read '#'
> >        wide character read 'i'
> > 
> >    Another simple example to reproduce the issue.:
> >    #include <stdio.h>
> >    #include <wchar.h>
> >    #include <stdlib.h> // for EXIT_SUCCESS
> >    #include <assert.h>
> > 
> >    int main() {
> >        enum { narrow = -1, query = 0, wide = 1 };
> >        FILE* test = fopen("test.bin", "r");
> >        if(!test) {
> >            puts("You need to have test.bin file.");
> >            abort();
> >        }
> >        // establish NARROW orientation
> >        fwide(test, narrow);
> >        // reopen to reset, broken on musl
> >        freopen("test.bin", "r", test);
> >        assert(fwide(test, query) == 0); // will fail on musl
> >        // CLEANUP
> >        fclose(test);
> >        // Exit the program
> >        return EXIT_SUCCESS;
> >    }
> > 
> >    -----
> >    Save as main.c, run with
> >    cc main.c
> >    echo test >test.bin
> >    ../a.out
> > 
> >    Will output
> >    Assertion failed: fwide(test, query) == 0 (main2.c: main: 17)
> >    Aborted
> 
> Indeed this looks like an oversight. I'll need to look at what it will
> take to fix it. Naively, just f->wide = 0 before successful return in
> freopen seems ok, but probably also f->locale needs to be set to 0 to
> reset the encoding rule, and it might be necessary to null out some
> buffer pointers to force future calls to stdio functions to reprobe
> the width.
> 
> Thanks for the report.
> 
> BTW is there a way you can fix the mailer you're using? The plain text
> version of your mail was not plain text, but was littered with html
> entities (&nbsp; etc.) making it unreadable, so I rendered the html
> version to text in order to be able to reply.

OK, fflush took care of the buffer pointers already, so just setting
f->mode and f->locale to 0 should be fine. I'll commit a fix.

Rich