[Toybox] freadahead()

[enh]

On Mon, Apr 11, 2022 at 3:19 PM Rob Landley <rob at landley.net> wrote:

> Elliott: could you add __freadahead() to bionic the way musl and z/os and
> so on
> have?
>
> I'm bothering the posix guys about this, thread starts here:
>
>   https://www.mail-archive.com/austin-group-l@opengroup.org/msg09468.html
>
> The problem is calling getline() then calling fileno() and reading from
> that can
> miss data in the FILE * buffer, and that's why patch.c is still using
> get_line()
> to do byte at a time reads so it can hand off to sendfile() when it runs
> out of
> hunks. It's also what wget wants for zlib support, it's likely to come up
> in
> Moritz' implementation of git, and so on. Basically anything with an
> smtp/mbox
> ancestry that needs a filehandle to process the "body" after the keyword:
> value
> headers.
>
> Unfortunately the ANSI C guys pointed out they don't do anything with file
> descriptors (no open() no read() no nothing), which is why fdopen() and
> fileno()
> (and for that matter getline()) live in posix, not ANSI C.
>
> But the Posix guys are so far saying (in private email, not to the list,
> sigh)
> that mucking about with the internals of an ANSI struct is thing that ANSI
> should do, not them. Except fileno() already does that, and ANSI hasn't got
> getline(), so the use case for this does not exist over there. You can ONLY
> access files through FILE * in pure ANSI C, so there's no context
> switching that
> can miss data...
>
> Anyway, if musl and bionic have __freadahead(), I can either #ifdef the
> glibc
> hack for it in portability.h, or might just move xgetline() into
> portability.c
> and have it do get_line() on systems that don't have __freadahead().


you're still screwed on macOS? (oh, except macOS seems to be just the
regular BSD header, with everything public and "roughly what you'd expect",
so that's easily done manually too.)

yeah, iirc i added everything that glibc had but didn't even look at musl.
(if you'd asked me even this morning, i'd have told you musl was a strict
subset of glibc. TIL.)


> Honestly, I
> have no sympathy for the glibc guys who implemented everything EXCEPT
> __freadahead() because they want to force people to use
> gnugnugnustallmanftaghnlib:
>

citation needed :-)

i'd be genuinely curious to know why glibc doesn't have this (and
especially "why musl does anyway"). i'll have to check what else is in
musl's <stdio_ext.h> that isn't in glibc's. looks like these:

size_t __freadahead(FILE *);
const char *__freadptr(FILE *, size_t *);
void __freadptrinc(FILE *, size_t);



>   https://linux.die.net/man/3/__fpending
>
> They have _fbufsize() but not how much of the buffer is used. That's just
> sad.
>

that does seem of questionable utility now you bring it up. (though as you
pointed out on the posix list, unless you _know_ you can seek on a stream,
all of this is pretty dodgy...)


> You can ask this question for output, but not for input. (And given that
> you
> can't plug the output any data you wrote might already have been sent so
> being
> able to "unsend" it without seeking is never reliable... sigh.)
>
> Rob
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.landley.net/pipermail/toybox-landley.net/attachments/20220411/6c893efa/attachment.htm>