[Toybox] [PATCH] Implement `file -`.

[Rob Landley]

On 05/05/2018 02:14 PM, enh wrote:
> On Sat, May 5, 2018 at 10:53 AM, Rob Landley <rob at landley.net> wrote:
> well, strictly i think only the ELF case actually needs this
> currently. we could fix the logic for the non-ELF cases (see below)
> and only copy to a temporary file for the ELF case. (alternatively, we
> could always read into memory and unify the cases that way --- from
> strace that seems to be what FSF file does?)

I prefer to have one codepath that works for everything and gets all the
testing. (The performance penalty was terrible enough I couldn't do it with
"tail" but that's why it's an exception. :)

>>> [PATCH] Implement `file -`.
>>>
>>> Previously we'd just always bogusly report "empty".
>>
>> That's not what I was seeing before this patch.
>>
>>   $ zcat ../filesystems.tar.gz | file -
>>   /dev/stdin: POSIX tar archive (GNU)
>>   $ zcat ../filesystems.tar.gz | ./file -
>>   -: POSIX tar archive (GNU)
>>
>> Even with input from a pipe, we're detecting stuff?
>>
>> Could you clarify what problem this patch is trying to solve? I don't understand.
> 
> i don't think you're accidentally running FSF file there, or you're
> accidentally running toybox file _with_ my patch.
> 
> here's a clean toybox checkout:

$ git clone https://github.com/landley/toybox walrus
Cloning into 'walrus'...
$ cd walrus
$ make defconfig
...
$ make
...

Ok, got a fresh copy, what are the tests...

> $ ./toybox file - < README
> -: empty
> $ ./toybox file - < toybox
> -: empty
> $ ./toybox file - < tests/files/tar/tar.tgz
> -: empty
> $ cat tests/files/tar/tar.tgz | ./toybox file -
> -: empty
> $ zcat tests/files/tar/tar.tgz | ./toybox file -
> -: empty
> $

Alright:

$ ./toybox file - < README
-: ASCII text

That's built against glibc, you're building against bionic?

$ ./toybox file - < toybox
-: file: mmap: Cannot allocate memory
ELF executable, 64-bit LSB x86-64$

Yeah, mmap is broken and the stdout/stderr out of sequence flushing produces a
screwed up shell prompt due to missing newline afterwards. *jazzhands*

$ ./toybox file - < ../filesystems.tar.gz
-: gzip compressed data
$ zcat ../filesystems.tar.gz | ./toybox file -
-: POSIX tar archive (GNU)

It looks like glibc is behaving very differently than what you're seeing.

Anything I build for x86-64 on the NDK segfaults before main() on the AMD C60 in
this netbook (although I ordered a new netbook from system76 today, which cost
the year the constitution was written _before_ shipping so maybe I'll have
better luck with that than the $200 thing from target I upgraded the RAM on)...

Let's try a static build against musl-libc...

$ zcat ../filesystems.tar.gz | ./toybox file -
-: empty
$ ./toybox file - < toybox
-: empty
$ ./toybox file - < README
-: empty

Huh. Ok, that's weird. Yes, glibc is doing something that other C libraries are
not doing, I'll see if I can track it down now that I can reproduce it...

Rob