Re: 900276502f..: Paul Eggert 2018-11-13 Act like POSIX sh if $HOME is relative

Re: 900276502f..: Paul Eggert 2018-11-13 Act like POSIX sh if $HOME is relative

[João Távora]

Hi,

I've only now caught up with the discussion of bug#33255.  Are the
consequences of interpreting the string "c:" on Windows as a relative
file name understood?

With HOME="z:", it used to be that (expand-file-name "~/blabla")
returned "z:/blabla" but now it returns
"z:/source/emacs/emacs/src/z:/blabla", which is not a valid Windows
pathname.

With HOME set thusly, this breaks a recent master build immediately (and
misteriously), since it can't find/create user-emacs-directory.

Notice however that (expand-file-name "z:") does return "z:/".  So I
think there is an inconsistency here, even if it is true that
(file-name-absolute-p "z:") has always returned nil.

All of this can be fixed by me by setting HOME to "z:/", but
unfortunately I am breaking other applications that expected it
slash-less.

An alternative would be:

* for this part of the change to be reverted in Emacs;

* for me to set the correct environment var from the Windows shortcut
  that points to runemacs.exe. Can it be done easily?

Anyway, maybe a tiny edit to the NEWS to mention this particular (and
presumably common) case could be worth it.

João Távora

Re: 900276502f..: Paul Eggert 2018-11-13 Act like POSIX sh if $HOME is relative

[Yuri Khan]

On Wed, Dec 12, 2018 at 7:43 PM João Távora <joaotavora@gmail.com> wrote:

> With HOME="z:", it used to be that (expand-file-name "~/blabla")
> returned "z:/blabla" but now it returns
> "z:/source/emacs/emacs/src/z:/blabla", which is not a valid Windows
> pathname.

As a former Windows user, I’d say both are wrong.

Setting HOME="z:" says “Whichever is the current directory on drive Z,
that’s my home”. ~/blabla means “file named blabla in my home”, i.e.
“blabla in the current directory on drive Z”, i.e. "z:blabla". Further
expansion depends on the actual current directory on Z:; if that’s
/source/emacs/emacs/src, then "z:/source/emacs/emacs/src/blabla".

> All of this can be fixed by me by setting HOME to "z:/", but
> unfortunately I am breaking other applications that expected it
> slash-less.

I’d say those are broken already. A slashless drive letter refers to
the current directory there, not the root.

> An alternative would be:

* get those other applications fixed;

> * for this part of the change to be reverted in Emacs;
>
> * for me to set the correct environment var from the Windows shortcut
>   that points to runemacs.exe. Can it be done easily?

Easiest is probably to change it to run a .cmd file that sets the
variable and then runs runemacs.exe, if you don’t mind a useless
console window. If you do, you’ll probably need to write and compile a
(GUI-targeted but windowless) program in a language of your choice,
that runs runemacs.exe with a modified environment.

Re: 900276502f..: Paul Eggert 2018-11-13 Act like POSIX sh if $HOME is relative

[João Távora]

[-- Attachment #1: Type: text/plain, Size: 987 bytes --]

 [Now ccing to emacs-devel]

On Wed, Dec 12, 2018 at 1:20 PM Yuri Khan <yurivkhan@gmail.com> wrote:

> An alternative would be:
>
> * get those other applications fixed;
>

Really? If you answer to a real-world problem is "fix the world" why even
bother answering?


>
> > * for this part of the change to be reverted in Emacs;
> >
> > * for me to set the correct environment var from the Windows shortcut
> >   that points to runemacs.exe. Can it be done easily?
>
> Easiest is probably to change it to run a .cmd file that sets the
> variable and then runs runemacs.exe, if you don’t mind a useless
> console window. If you do, you’ll probably need to write and compile a
> (GUI-targeted but windowless) program in a language of your choice,
> that runs runemacs.exe with a modified environment.
>

At which point I might as well fix it in my Emacs config, which has ample
facilities for this, and which I would do if this weren't such an early
thing.

[-- Attachment #2: Type: text/html, Size: 1693 bytes --]

Re: 900276502f..: Paul Eggert 2018-11-13 Act like POSIX sh if $HOME is relative

[Eli Zaretskii]

> From: Yuri Khan <yurivkhan@gmail.com>
> Date: Wed, 12 Dec 2018 20:19:55 +0700
> Cc: Emacs developers <emacs-devel@gnu.org>, Paul Eggert <eggert@cs.ucla.edu>, 
> 	Eli Zaretskii <eliz@gnu.org>
> 
> On Wed, Dec 12, 2018 at 7:43 PM João Távora <joaotavora@gmail.com> wrote:
> 
> > With HOME="z:", it used to be that (expand-file-name "~/blabla")
> > returned "z:/blabla" but now it returns
> > "z:/source/emacs/emacs/src/z:/blabla", which is not a valid Windows
> > pathname.
> 
> As a former Windows user, I’d say both are wrong.
> 
> Setting HOME="z:" says “Whichever is the current directory on drive Z

Emacs doesn't support drive-relative file names on Windows.  It never
did.  It always interpreted "x:" as the root directory on drive x.

RE: 900276502f..: Paul Eggert 2018-11-13 Act like POSIX sh if $HOME is relative

[Drew Adams]

> > Setting HOME="z:" says “Whichever is the current directory on drive Z
> 
> Emacs doesn't support drive-relative file names on Windows.  It never
> did.  It always interpreted "x:" as the root directory on drive x.

Whatever gets decided wrt this bug report, please ensure
that the Emacs docs make clear what the behavior is.

In particular, let's tell users that "Emacs doesn't
support drive-relative file names on Windows."  So,
e.g., it interprets "x:" as the root directory on
drive x.

(Perhaps the behavior is already documented, in which
case no change is needed - dunno.)

Re: 900276502f..: Paul Eggert 2018-11-13 Act like POSIX sh if $HOME is relative

[Eli Zaretskii]

> Date: Wed, 12 Dec 2018 07:29:48 -0800 (PST)
> From: Drew Adams <drew.adams@oracle.com>
> Cc: eggert@cs.ucla.edu, joaotavora@gmail.com, emacs-devel@gnu.org
> 
> > > Setting HOME="z:" says “Whichever is the current directory on drive Z
> > 
> > Emacs doesn't support drive-relative file names on Windows.  It never
> > did.  It always interpreted "x:" as the root directory on drive x.
> 
> Whatever gets decided wrt this bug report, please ensure
> that the Emacs docs make clear what the behavior is.
> 
> In particular, let's tell users that "Emacs doesn't
> support drive-relative file names on Windows."  So,
> e.g., it interprets "x:" as the root directory on
> drive x.

We didn't explain this subtlety before, and I don't see why explain it
now.  It's a corner use case (I'd actually consider it a bug in the
user configuration).  Since there's no change in behavior here, we
don't need to change anything in the documentation.

Re: 900276502f..: Paul Eggert 2018-11-13 Act like POSIX sh if $HOME is relative

[Óscar Fuentes]

"João Távora" <joaotavora@gmail.com> writes:

> I've only now caught up with the discussion of bug#33255.  Are the
> consequences of interpreting the string "c:" on Windows as a relative
> file name understood?
>
> With HOME="z:", it used to be that (expand-file-name "~/blabla")
> returned "z:/blabla" but now it returns
> "z:/source/emacs/emacs/src/z:/blabla", which is not a valid Windows
> pathname.

[snip]

I was about to post a reply but Yuri already answered, so I'll only add
a pointer to an official reference about this topic:

https://docs.microsoft.com/en-us/dotnet/standard/io/file-path-formats

Re: 900276502f..: Paul Eggert 2018-11-13 Act like POSIX sh if $HOME is relative

[João Távora]

[-- Attachment #1: Type: text/plain, Size: 1132 bytes --]

On Wed, Dec 12, 2018 at 1:41 PM Óscar Fuentes <ofv@wanadoo.es> wrote:

> "João Távora" <joaotavora@gmail.com> writes:
>
> > I've only now caught up with the discussion of bug#33255.  Are the
> > consequences of interpreting the string "c:" on Windows as a relative
> > file name understood?
> >
> > With HOME="z:", it used to be that (expand-file-name "~/blabla")
> > returned "z:/blabla" but now it returns
> > "z:/source/emacs/emacs/src/z:/blabla", which is not a valid Windows
> > pathname.
>
> [snip]
>
> I was about to post a reply but Yuri already answered, so I'll only add
> a pointer to an official reference about this topic:
>
> https://docs.microsoft.com/en-us/dotnet/standard/io/file-path-formats
>

Óscar, I get it that z:bla has a meaning. But is this meaning ever useful
when for expanding "~/bla"?

I didn't say the change was "wrong".  I just asked, basically:

* if anyone had contemplated this particular problem on Windows;
* if it's possible to interpret HOME specially for this particular bit of
backward compatibility without breaking the cases of bug#33255

João Távora

[-- Attachment #2: Type: text/html, Size: 1876 bytes --]

Re: 900276502f..: Paul Eggert 2018-11-13 Act like POSIX sh if $HOME is relative

[Óscar Fuentes]

João Távora <joaotavora@gmail.com> writes:

> On Wed, Dec 12, 2018 at 1:41 PM Óscar Fuentes <ofv@wanadoo.es> wrote:
>
>> "João Távora" <joaotavora@gmail.com> writes:
>>
>> > I've only now caught up with the discussion of bug#33255.  Are the
>> > consequences of interpreting the string "c:" on Windows as a relative
>> > file name understood?
>> >
>> > With HOME="z:", it used to be that (expand-file-name "~/blabla")
>> > returned "z:/blabla" but now it returns
>> > "z:/source/emacs/emacs/src/z:/blabla", which is not a valid Windows
>> > pathname.
>>
>> [snip]
>>
>> I was about to post a reply but Yuri already answered, so I'll only add
>> a pointer to an official reference about this topic:
>>
>> https://docs.microsoft.com/en-us/dotnet/standard/io/file-path-formats
>>
>
> Óscar, I get it that z:bla has a meaning. But is this meaning ever useful
> when for expanding "~/bla"?
>
> I didn't say the change was "wrong".  I just asked, basically:
>
> * if anyone had contemplated this particular problem on Windows;
> * if it's possible to interpret HOME specially for this particular bit of
> backward compatibility without breaking the cases of bug#33255

I don't know about the motivation of this change and have no time to
study it (a cursory look at the diff shows some concerning points wrt
Windows compatibility such as

        hope one's home directory doesn't contain ':' or '%'.  */

).

A Windows path with no slash after the volume letter is relative and
always was. For the specific case of HOME, I can't think of a sane
scenario where a user could take advantage of interpreting V: as a
relative path for HOME on Emacs, but maybe I'm not being imaginative
enough.

You are asking for being bug-compatible with previous Emacs because you
must cope with bugs on other applications :-) Do you think that this
problem is frequent enough? What applications are those that don't admit
an slash after the volume letter in HOME?

Re: 900276502f..: Paul Eggert 2018-11-13 Act like POSIX sh if $HOME is relative

[João Távora]

[-- Attachment #1: Type: text/plain, Size: 657 bytes --]

On Wed, Dec 12, 2018 at 3:15 PM Óscar Fuentes <ofv@wanadoo.es> wrote:

>
>
> You are asking for being bug-compatible with previous Emacs because you
> must cope with bugs on other applications :-)


Yes, if you will.  That is what the world is like.



> Do you think that this
> problem is frequent enough?


Every time I start Emacs, or everytime someone in my $DAYJOB starts Emacs
after upgrading to Emacs 27.  Frequent enough?



> What applications are those that don't admit
> an slash after the volume letter in HOME?
>

30-year+ old, big bad misbehaving Lisp-based applications.  Kinda like
Emacs, I suppose :-)

João

[-- Attachment #2: Type: text/html, Size: 1250 bytes --]

Re: 900276502f..: Paul Eggert 2018-11-13 Act like POSIX sh if $HOME is relative

[Óscar Fuentes]

João Távora <joaotavora@gmail.com> writes:

> On Wed, Dec 12, 2018 at 3:15 PM Óscar Fuentes <ofv@wanadoo.es> wrote:
>
>>
>>
>> You are asking for being bug-compatible with previous Emacs because you
>> must cope with bugs on other applications :-)
>
>
> Yes, if you will.  That is what the world is like.

Yes, sadly :-)

>> Do you think that this
>> problem is frequent enough?
>
>
> Every time I start Emacs, or everytime someone in my $DAYJOB starts Emacs
> after upgrading to Emacs 27.  Frequent enough?

I was asking about an heuristic to estimate the hypothetical fraction of
affected users. Something like "I know two guys on the Internet with the
same problem."

>> What applications are those that don't admit
>> an slash after the volume letter in HOME?
>>
>
> 30-year+ old, big bad misbehaving Lisp-based applications.  Kinda like
> Emacs, I suppose :-)

Let me guess: those applications share the same runtime and/or
libraries.

FWIW, I'm not opposed to keeping Emacs bug-compatible on this case. (As
if my opinion had some value :-)

Re: 900276502f..: Paul Eggert 2018-11-13 Act like POSIX sh if $HOME is relative

[João Távora]

[-- Attachment #1: Type: text/plain, Size: 984 bytes --]

On Wed, Dec 12, 2018 at 4:56 PM Óscar Fuentes <ofv@wanadoo.es> wrote:

> > Every time I start Emacs, or everytime someone in my $DAYJOB starts Emacs
> > after upgrading to Emacs 27.  Frequent enough?
>
> I was asking about an heuristic to estimate the hypothetical fraction of
> affected users. Something like "I know two guys on the Internet with the
> same problem."
>

OK. So it's more like "I know 50 guys/gals in the same building."


> >> What applications are those that don't admit
> >> an slash after the volume letter in HOME?
> >>
> >
> > 30-year+ old, big bad misbehaving Lisp-based applications.  Kinda like
> > Emacs, I suppose :-)
>
> Let me guess: those applications share the same runtime and/or
> libraries.
>

Nope. No shared libraries here. Wish there were, tbh, but nope.

We'd all sooner be told to start changing environment vars furiously
(or never upgrade to Emacs 27) than fixing and rebuilding every
such application.

João

[-- Attachment #2: Type: text/html, Size: 1547 bytes --]

Re: 900276502f..: Paul Eggert 2018-11-13 Act like POSIX sh if $HOME is relative

[Eli Zaretskii]

> From: Óscar Fuentes <ofv@wanadoo.es>
> Date: Wed, 12 Dec 2018 17:55:12 +0100
> 
> (As if my opinion had some value :-)

It does.

Thanks.

Re: 900276502f..: Paul Eggert 2018-11-13 Act like POSIX sh if $HOME is relative

[Stefan Monnier]

> I was about to post a reply but Yuri already answered, so I'll only add
> a pointer to an official reference about this topic:
> https://docs.microsoft.com/en-us/dotnet/standard/io/file-path-formats

Emacs only has `default-directory` to keep track of the "current
directory" so it has no notion of having a different "current directory"
on every "drive".

So I think a $HOME of "z:" should be converted early on to
"z:/<something>" and after that, "p:" can only be interpreted by Emacs
as "p:/"


        Stefan

Re: 900276502f..: Paul Eggert 2018-11-13 Act like POSIX sh if $HOME is relative

[Eli Zaretskii]

> From: Stefan Monnier <monnier@iro.umontreal.ca>
> Date: Wed, 12 Dec 2018 10:16:14 -0500
> 
> > I was about to post a reply but Yuri already answered, so I'll only add
> > a pointer to an official reference about this topic:
> > https://docs.microsoft.com/en-us/dotnet/standard/io/file-path-formats
> 
> Emacs only has `default-directory` to keep track of the "current
> directory" so it has no notion of having a different "current directory"
> on every "drive".

Emacs doesn't, but some systems do.

> So I think a $HOME of "z:" should be converted early on to
> "z:/<something>" and after that, "p:" can only be interpreted by Emacs
> as "p:/"

That's what Emacs was always doing on Windows, but not on DOS.  I see
no reason to change this behavior, it doesn't contradict the changes
we did for the HOME directory being a relative file name.

Re: 900276502f..: Paul Eggert 2018-11-13 Act like POSIX sh if $HOME is relative

[Eli Zaretskii]

> From: "João Távora" <joaotavora@gmail.com>
> Date: Wed, 12 Dec 2018 12:42:30 +0000
> 
> With HOME="z:", it used to be that (expand-file-name "~/blabla")
> returned "z:/blabla" but now it returns
> "z:/source/emacs/emacs/src/z:/blabla", which is not a valid Windows
> pathname.

That's a regression that should be fixed.

> All of this can be fixed by me by setting HOME to "z:/", but
> unfortunately I am breaking other applications that expected it
> slash-less.

Those other applications have a bug as well, so I urge you to fix that
or find some fixed version of them.

> An alternative would be:

I see no reason for alternatives, we just need to fix the regression.

Re: 900276502f..: Paul Eggert 2018-11-13 Act like POSIX sh if $HOME is relative

[João Távora]

[-- Attachment #1: Type: text/plain, Size: 233 bytes --]

On Wed, Dec 12, 2018 at 3:20 PM Eli Zaretskii <eliz@gnu.org> wrote:

>
> I see no reason for alternatives, we just need to fix the regression.
>

Cool! (Tho one of the alternatives really was "fix the regression" :-)

João

[-- Attachment #2: Type: text/html, Size: 505 bytes --]

Re: 900276502f..: Paul Eggert 2018-11-13 Act like POSIX sh if $HOME is relative

[Eli Zaretskii]

> From: João Távora <joaotavora@gmail.com>
> Date: Wed, 12 Dec 2018 15:30:54 +0000
> Cc: emacs-devel <emacs-devel@gnu.org>, Paul Eggert <eggert@cs.ucla.edu>
> 
> On Wed, Dec 12, 2018 at 3:20 PM Eli Zaretskii <eliz@gnu.org> wrote:
> 
>  I see no reason for alternatives, we just need to fix the regression.
> 
> Cool! (Tho one of the alternatives really was "fix the regression" :-) 

Should be fixed now.