Re: zsh: RM_STAR warning counts files as though DOT_GLOB is set

Re: zsh: RM_STAR warning counts files as though DOT_GLOB is set

[Daniel Shahaf]

Forwarding from Debian:

Daniel Shahaf wrote on Mon, 11 Sep 2017 15:07 +0000:
> $ zsh -f
> % cd $(mktemp -d)
> % touch ./foo ./.bar 
> % rm *
> zsh: sure you want to delete all 2 files in /tmp/tmp.65xnqgzMAY [yn]? y
> % 
> % ls -A
> .bar
> % echo $ZSH_PATCHLEVEL
> zsh-5.4.2-1-gab6ac7313

The problem is that it says "2 files" but only one file is deleted.

The impedance mismatch is that checkrmall() calls zreaddir(...,
unset(GLOBDOTS)), but the second actual argument to zreaddir() governs
whether the dot and dot-dot entries would be excluded.  Instead,
checkrmall() should pass true for the second argument of zreaddir(),
and dotfiles should be ignored by some other means when ignoredots is true.
(I've just added a docstring to zreaddir() to close the barn door.)

I don't understand the last branch of the if/else in that function: it seems to
prompt "... delete all the files ..." when count == 0?

Cheers,

Daniel

Re: zsh: RM_STAR warning counts files as though DOT_GLOB is set

[Peter Stephenson]

On Mon, 11 Sep 2017 15:19:15 +0000
Daniel Shahaf <danielsh@apache.org> wrote:
> I don't understand the last branch of the if/else in that function: it
> seems to prompt "... delete all the files ..." when count == 0?

Apparently, yes.  I think it's just covering the bases.  I haven't
looked, but probably the stuff above it gradually expanded to make this
case logically redundant, unless there's a mismatch.

pws

Re: zsh: RM_STAR warning counts files as though DOT_GLOB is set

[Bart Schaefer]

On Mon, Sep 11, 2017 at 8:23 AM, Peter Stephenson
<p.stephenson@samsung.com> wrote:
> On Mon, 11 Sep 2017 15:19:15 +0000
> Daniel Shahaf <danielsh@apache.org> wrote:
>> I don't understand the last branch of the if/else in that function: it
>> seems to prompt "... delete all the files ..." when count == 0?
>
> Apparently, yes.  I think it's just covering the bases.

The issue here, as I recall, is that it's possible to have:
1 "rm *" is issued
2 chkrmall() finds no files in the directory
3 some external process creates a file
4 "rm *" is executed
5 the file created at (3) is silently deleted

So we would need either to avoid re-expanding "*" after chkrmall(), or
treat "found no files" as "user answered no".

Re: zsh: RM_STAR warning counts files as though DOT_GLOB is set

[Daniel Shahaf]

Bart Schaefer wrote on Mon, 11 Sep 2017 08:31 -0700:
> On Mon, Sep 11, 2017 at 8:23 AM, Peter Stephenson
> <p.stephenson@samsung.com> wrote:
> > On Mon, 11 Sep 2017 15:19:15 +0000
> > Daniel Shahaf <danielsh@apache.org> wrote:
> >> I don't understand the last branch of the if/else in that function: it
> >> seems to prompt "... delete all the files ..." when count == 0?
> >
> > Apparently, yes.  I think it's just covering the bases.
> 
> The issue here, as I recall, is that it's possible to have:
> 1 "rm *" is issued
> 2 chkrmall() finds no files in the directory
> 3 some external process creates a file
> 4 "rm *" is executed
> 5 the file created at (3) is silently deleted
> 
> So we would need either to avoid re-expanding "*" after chkrmall(), or
> treat "found no files" as "user answered no".

Good catch.

The warning messages emitted by the "if" and "else if" branches are all
susceptible to this race condition.  They were all added in 37946 in
order to "make [the] warnings more informative".

Simply reverting 37946 would resolve both the OP's problem and this race
condition. (Just stating a fact; maybe there are other/better solutions)

Cheers,

Daniel

Re: zsh: RM_STAR warning counts files as though DOT_GLOB is set

[Bart Schaefer]

On Sep 11,  3:41pm, Daniel Shahaf wrote:
}
} The warning messages emitted by the "if" and "else if" branches are all
} susceptible to this race condition.  They were all added in 37946 in
} order to "make [the] warnings more informative".

IIRC this was discussed at the time, and the conclusion was that the
race condition is acceptable as long as the user is prompted and must
answer yes -- the count needn't be accurate when non-zero, it just
mustn't be skipped when zero.