[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
bug#46284: 27.1; emacs-27: windows-nt regression with process sentinel's
From: |
Ioannis Kappas |
Subject: |
bug#46284: 27.1; emacs-27: windows-nt regression with process sentinel's change description argument Previous Next |
Date: |
Thu, 4 Feb 2021 18:12:06 +0000 |
Looks good, passes on both the original ert test case I’ve attached to the bug
report and the particular case I’ve encountered the issue on (bringing down a
REPL with cider).
> On 4 Feb 2021, at 16:18, Eli Zaretskii <eliz@gnu.org> wrote:
>
>> From: Ioannis Kappas <ioannis.kappas@gmail.com>
>> Date: Thu, 4 Feb 2021 07:36:23 +0000
>>
>> It appears as if the underlying signal number to description conversion
>> method is broken in emacs-27 on windows-nt causing the regression.
>>
>> Looking at the code, it looks like the issue is around the area which
>> attempts to convert a signal number to a string description. On emacs
>> version prior and including emacs-27, the conversion is done by
>> directly accessing the _sys_siglist[] table provided in the unix
>> systems, mapping signal numbers to signal descriptions, e.g. SIGINT ->
>> "Interrupted". On native windows though, this table does not exist, and
>> emacs simulates it in src/sysdep.c:init_signals() so as to conform with
>> the rest of the code which expects this table to be there.
>>
>> init_signals() only populates the descriptions in the C table when emacs is
>> not
>> !initialized, i.e. during the dumping phase. When emacs is thus ran
>> normally, it expects this table to have been loaded from the dump into
>> memory.
>>
>> This table though does not appear to make it through to the dump in
>> emacs-27. Having a look at the new pdumper, it looks like that it
>> performs differently than its predecessor. It seems as if it
>> only cover lisp constructs, while Unexec was also dumping the data
>> section of the process from memory? If true then, it implies that this
>> table is not eligible for dumping any more (since it is a C array), but
>> should always be
>> initialised when emacs is invoked by the user. This has a very
>> simple solution, taking out the if (!initialized) line in
>> src/sysdep.c:init_signals():
>
> Thanks for the analysis and the patch. I'd prefer to fix this in a
> slightly different way, which keeps the support for building Emacs
> with unexec. Does the following patch work for you?
>
> diff --git a/src/sysdep.c b/src/sysdep.c
> index f94ce4d..d100a5c 100644
> --- a/src/sysdep.c
> +++ b/src/sysdep.c
> @@ -1980,7 +1980,8 @@ init_signals (void)
> #endif
>
> #if !HAVE_DECL_SYS_SIGLIST && !defined _sys_siglist
> - if (! initialized)
> + if (! initialized
> + || dumped_with_pdumper_p ())
> {
> sys_siglist[SIGABRT] = "Aborted";
> # ifdef SIGAIO