bug#18370: insert-file-contents: forbids also beg, end for non-regular (special) files

bug#18370: insert-file-contents: forbids also beg, end for non-regular (special) files

[Ivan Shmakov]

Package:  emacs
Severity: minor

	As currently implemented, insert-file-contents disallows not
	only (as per (elisp.info) Reading from Files) the case of either
	‘visit’ or ‘replace’ arguments being non-nil, but also when
	non-nil are either ‘beg’ or ‘end’ (or both):

  3506	  /* This code will need to be changed in order to work on named
  3507	     pipes, and it's probably just not worth it.  So we should at
  3508	     least signal an error.  */
  3509	  if (!S_ISREG (st.st_mode))
  3510	    {
  3511	      not_regular = 1;
⋯
  3516	      if (! NILP (replace) || ! NILP (beg) || ! NILP (end))
  3517		xsignal2 (Qfile_error,
  3518			  build_string ("not a regular file"), orig_filename);
  3519	    }

	This, however, precludes the use of insert-file-contents not
	only on named pipes, but also on /dev/cdrom, /dev/random, and
	the like, – for (and especially in the case of the latter) the
	‘end’ argument gets rather essential here, as otherwise the
	function is likely to read much more than the caller will be
	able to handle at any single time.

	Instead, I’d suggest that ‘end’ is always allowed, and ‘beg’ is
	allowed when the file in question is /seekable/, – which, ISTR,
	is possible to check beforehand (lseek (fd, 0, SEEK_CUR) < 0?)

	From a glance over the code, this new behavior wouldn’t be all
	that hard to implement (some not_regular checks will have to be
	replaced with the ones against beg_offset, end_offset, etc.),
	but from what I read, – it’s already going to be more than I can
	test right now.  So no .diff this time, alas.

-- 
FSF associate member #7257  http://boycottsystemd.org/  … 3013 B6A0 230E 334A

bug#18370: insert-file-contents: forbids also beg, end for non-regular (special) files

[Lars Ingebrigtsen]

Ivan Shmakov <ivan@siamics.net> writes:

> 	As currently implemented, insert-file-contents disallows not
> 	only (as per (elisp.info) Reading from Files) the case of either
> 	‘visit’ or ‘replace’ arguments being non-nil, but also when
> 	non-nil are either ‘beg’ or ‘end’ (or both):

[...]

> 	This, however, precludes the use of insert-file-contents not
> 	only on named pipes, but also on /dev/cdrom, /dev/random, and
> 	the like, – for (and especially in the case of the latter) the
> 	‘end’ argument gets rather essential here, as otherwise the
> 	function is likely to read much more than the caller will be
> 	able to handle at any single time.
>
> 	Instead, I’d suggest that ‘end’ is always allowed, and ‘beg’ is
> 	allowed when the file in question is /seekable/, – which, ISTR,
> 	is possible to check beforehand (lseek (fd, 0, SEEK_CUR) < 0?)
>
> 	From a glance over the code, this new behavior wouldn’t be all
> 	that hard to implement (some not_regular checks will have to be
> 	replaced with the ones against beg_offset, end_offset, etc.),
> 	but from what I read, – it’s already going to be more than I can
> 	test right now.  So no .diff this time, alas.

(I'm going through old bug reports that unfortunately weren't resolved
at the time.)

Simple test case:

(insert-file-contents "/dev/urandom" nil nil 10)

This still fails in Emacs 29, and it would indeed be nice if it worked.
I seem to remember this being discussed previously, and that...  er...
no, I can't really recall anything more than that.

Does anybody see any good reasons why we shouldn't allow this?  If not,
I can take a stab at implementing it...

-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no

bug#18370: insert-file-contents: forbids also beg, end for non-regular (special) files

[Lars Ingebrigtsen]

Lars Ingebrigtsen <larsi@gnus.org> writes:

> Simple test case:
>
> (insert-file-contents "/dev/urandom" nil nil 10)
>
> This still fails in Emacs 29, and it would indeed be nice if it worked.
> I seem to remember this being discussed previously, and that...  er...
> no, I can't really recall anything more than that.
>
> Does anybody see any good reasons why we shouldn't allow this?  If not,
> I can take a stab at implementing it...

There were no objections, so I've now added this to Emacs 29.

-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no

bug#18370: insert-file-contents: forbids also beg, end for non-regular (special) files

[Juri Linkov]

>> Simple test case:
>>
>> (insert-file-contents "/dev/urandom" nil nil 10)
>>
>> This still fails in Emacs 29, and it would indeed be nice if it worked.
>> I seem to remember this being discussed previously, and that...  er...
>> no, I can't really recall anything more than that.
>>
>> Does anybody see any good reasons why we shouldn't allow this?  If not,
>> I can take a stab at implementing it...
>
> There were no objections, so I've now added this to Emacs 29.

Maybe this is related to bug#9800, still unfixed after this change.

bug#18370: insert-file-contents: forbids also beg, end for non-regular (special) files

[Lars Ingebrigtsen]

Juri Linkov <juri@linkov.net> writes:

> Maybe this is related to bug#9800, still unfixed after this change.

I think it's somewhat unrelated -- the /proc files are "regular" files,
they just tend to "grow" while reading.  I think the solution sketched
in that bug report sounds OK.

-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no