Re: [Emacs-diffs] master 1072155: Avoid duplicate entries in process-environment after re-dumping

Re: [Emacs-diffs] master 1072155: Avoid duplicate entries in process-environment after re-dumping

[Daniel Colascione]

> branch: master
> commit 107215596c1a8edfb239a88850d822642bc0e4af
> Author: Eli Zaretskii <eliz@gnu.org>
> Commit: Eli Zaretskii <eliz@gnu.org>
>
>     Avoid duplicate entries in process-environment after re-dumping
>
>     * src/pdumper.c (Fdump_emacs_portable): Reset
>     process-environment to nil.  (Bug#34936)
> ---
>  src/pdumper.c | 6 ++++++
>  1 file changed, 6 insertions(+)
>
> diff --git a/src/pdumper.c b/src/pdumper.c
> index fbf17d1..f459d97 100644
> --- a/src/pdumper.c
> +++ b/src/pdumper.c
> @@ -4025,6 +4025,12 @@ types.  */)
>    Lisp_Object symbol = intern ("command-line-processed");
>    specbind (symbol, Qnil);
>
> +  /* Reset process-environment -- this is for when they re-dump a
> +     pdump-restored emacs, since set_initial_environment wants always
> +     to cons it from scratch.  */
> +  Vprocess_environment = Qnil;

Don't we want to reset process-environment to its old value in
dump_unwind_cleanup?

> +  garbage_collect ();
> +
>    CHECK_STRING (filename);
>    filename = Fexpand_file_name (filename, Qnil);
>    filename = ENCODE_FILE (filename);

Does it make sense to move this chunk before the
garbage-collect-until-we-run-all-finalizers loop above? That way, we'd run
one fewer GC.

Re: [Emacs-diffs] master 1072155: Avoid duplicate entries in process-environment after re-dumping

[Eli Zaretskii]

> Date: Tue, 2 Apr 2019 12:02:55 -0700
> From: "Daniel Colascione" <dancol@dancol.org>
> 
> > +  /* Reset process-environment -- this is for when they re-dump a
> > +     pdump-restored emacs, since set_initial_environment wants always
> > +     to cons it from scratch.  */
> > +  Vprocess_environment = Qnil;
> 
> Don't we want to reset process-environment to its old value in
> dump_unwind_cleanup?

You are thinking about re-dumping from an interactive session?  In
that case, probably yes.  But currently we only support dumping from
batch sessions, and in that case I don't see a need to restore
process-environment, am I missing something?

> > +  garbage_collect ();
> > +
> >    CHECK_STRING (filename);
> >    filename = Fexpand_file_name (filename, Qnil);
> >    filename = ENCODE_FILE (filename);
> 
> Does it make sense to move this chunk before the
> garbage-collect-until-we-run-all-finalizers loop above? That way, we'd run
> one fewer GC.

Probably.  I just wanted to do this as close to the actual dumping as
possible, but maybe it's not that important.

Re: [Emacs-diffs] master 1072155: Avoid duplicate entries in process-environment after re-dumping

[Daniel Colascione]

>> Date: Tue, 2 Apr 2019 12:02:55 -0700
>> From: "Daniel Colascione" <dancol@dancol.org>
>>
>> > +  /* Reset process-environment -- this is for when they re-dump a
>> > +     pdump-restored emacs, since set_initial_environment wants always
>> > +     to cons it from scratch.  */
>> > +  Vprocess_environment = Qnil;
>>
>> Don't we want to reset process-environment to its old value in
>> dump_unwind_cleanup?
>
> You are thinking about re-dumping from an interactive session?

Yes

> In
> that case, probably yes.  But currently we only support dumping from
> batch sessions, and in that case I don't see a need to restore
> process-environment, am I missing something?

We "support" dumping only from dedicated batch instances in that for now
we should consider only bugs in that use case release-blocking, but I
don't want to gratuitously break other use cases like dumping interactive
sessions without killing them, since these use cases are meant to work and
we'll give them the same level of support someday soon.

>
>> > +  garbage_collect ();
>> > +
>> >    CHECK_STRING (filename);
>> >    filename = Fexpand_file_name (filename, Qnil);
>> >    filename = ENCODE_FILE (filename);
>>
>> Does it make sense to move this chunk before the
>> garbage-collect-until-we-run-all-finalizers loop above? That way, we'd
>> run
>> one fewer GC.
>
> Probably.  I just wanted to do this as close to the actual dumping as
> possible, but maybe it's not that important.
>
>

Re: [Emacs-diffs] master 1072155: Avoid duplicate entries in process-environment after re-dumping

[Eli Zaretskii]

> Date: Tue, 2 Apr 2019 13:11:53 -0700
> From: "Daniel Colascione" <dancol@dancol.org>
> Cc: "Daniel Colascione" <dancol@dancol.org>,
>  emacs-devel@gnu.org
> 
> >> Don't we want to reset process-environment to its old value in
> >> dump_unwind_cleanup?
> >
> > You are thinking about re-dumping from an interactive session?
> 
> Yes
> 
> > In
> > that case, probably yes.  But currently we only support dumping from
> > batch sessions, and in that case I don't see a need to restore
> > process-environment, am I missing something?
> 
> We "support" dumping only from dedicated batch instances in that for now
> we should consider only bugs in that use case release-blocking, but I
> don't want to gratuitously break other use cases like dumping interactive
> sessions without killing them, since these use cases are meant to work and
> we'll give them the same level of support someday soon.

OK, but is there any reason dump_unwind_cleanup doesn't restore
post-gc-hook?  Omission?

Re: [Emacs-diffs] master 1072155: Avoid duplicate entries in process-environment after re-dumping

[Daniel Colascione]

>> Date: Tue, 2 Apr 2019 13:11:53 -0700
>> From: "Daniel Colascione" <dancol@dancol.org>
>> Cc: "Daniel Colascione" <dancol@dancol.org>,
>>  emacs-devel@gnu.org
>>
>> >> Don't we want to reset process-environment to its old value in
>> >> dump_unwind_cleanup?
>> >
>> > You are thinking about re-dumping from an interactive session?
>>
>> Yes
>>
>> > In
>> > that case, probably yes.  But currently we only support dumping from
>> > batch sessions, and in that case I don't see a need to restore
>> > process-environment, am I missing something?
>>
>> We "support" dumping only from dedicated batch instances in that for now
>> we should consider only bugs in that use case release-blocking, but I
>> don't want to gratuitously break other use cases like dumping
>> interactive
>> sessions without killing them, since these use cases are meant to work
>> and
>> we'll give them the same level of support someday soon.
>
> OK, but is there any reason dump_unwind_cleanup doesn't restore
> post-gc-hook?  Omission?

Not restoring post-gc-hook is a bug. Thanks for spotting that!

Re: [Emacs-diffs] master 1072155: Avoid duplicate entries in process-environment after re-dumping

[Eli Zaretskii]

> Date: Wed, 3 Apr 2019 09:58:35 -0700
> From: "Daniel Colascione" <dancol@dancol.org>
> Cc: "Daniel Colascione" <dancol@dancol.org>,
>  emacs-devel@gnu.org
> 
> >> We "support" dumping only from dedicated batch instances in that for now
> >> we should consider only bugs in that use case release-blocking, but I
> >> don't want to gratuitously break other use cases like dumping
> >> interactive
> >> sessions without killing them, since these use cases are meant to work
> >> and
> >> we'll give them the same level of support someday soon.
> >
> > OK, but is there any reason dump_unwind_cleanup doesn't restore
> > post-gc-hook?  Omission?
> 
> Not restoring post-gc-hook is a bug. Thanks for spotting that!

Thanks, I fixed both issues.