Re: Bug#99095: Process completion for gdb

Re: Bug#99095: Process completion for gdb

[Clint Adams]

> This works for me.  In fact, I've set:
> 
> zstyle ':completion:*:complete:*:*:processes' command 'ps ux'
> 
> Since this is usually what I want (and the information from 'ps u' makes it
> easier to find the desired process)
> 
> For gdb specifically, this isn't as useful as it might be.  Since filenames are
> completed as well, the process list will often scroll off the screen when run
> from a directory full of source code and object files.  Maybe the filename
> match should be restricted to '{,*.}core'?

We could do this to _gdb

50c50
<     _alternative 'files:: _files' "processes:: _pids -m ${w[1]:t}"
---
>     _alternative "core-files:: _files -g '*core'" 'files:: _files' "processes:: _pids -m ${w[1]:t}"

and then you could set zstyle ':completion:*:complete:gdb:*' tag-order 'processes core-files'

What do people think?

> Or perhaps to have some portable and extensible mechanism for selecting
> processes.  I would say that at least these should be supported:
> 
> - All of a user's processes (ps x, ps -u user), for most uses
> - All of everyone's processes (ps ax, ps -e), for root
> - Some formatting options (ps u, ps -f) for more verbosity

The problem is accurately determining whether ps is SYSV- or BSD-style or
something else entirely.

Re: Bug#99095: Process completion for gdb

[Sven Wischnowsky]

Clint Adams wrote:

> ...
> 
> We could do this to _gdb
> 
> 50c50
> <     _alternative 'files:: _files' "processes:: _pids -m ${w[1]:t}"
> ---
> >     _alternative "core-files:: _files -g '*core'" 'files:: _files' "processes:: _pids -m ${w[1]:t}"
> 
> and then you could set zstyle ':completion:*:complete:gdb:*' tag-order 'processes core-files'
> 
> What do people think?

I thought about something like that when I wrote _gdb.  The problem is
if there is no file matching *core.  Then the first would complete all
files (or directories), too.  And since they may be in different groups,
they would be shown twice.

Obviously, I didn't find a solution...

> > Or perhaps to have some portable and extensible mechanism for selecting
> > processes.  I would say that at least these should be supported:
> > 
> > - All of a user's processes (ps x, ps -u user), for most uses
> > - All of everyone's processes (ps ax, ps -e), for root
> > - Some formatting options (ps u, ps -f) for more verbosity
> 
> The problem is accurately determining whether ps is SYSV- or BSD-style or
> something else entirely.

Indeed.


Bye
  Sven


-- 
Sven Wischnowsky                         wischnow@informatik.hu-berlin.de

Re: Bug#99095: Process completion for gdb

[Clint Adams]

> I thought about something like that when I wrote _gdb.  The problem is
> if there is no file matching *core.  Then the first would complete all
> files (or directories), too.  And since they may be in different groups,
> they would be shown twice.
> 
> Obviously, I didn't find a solution...

One could rely on the output of file(1), but that's ridiculously expensive.
Why not an option to _path_files that doesn't add any matches if the
pattern is not met?

Re: Bug#99095: Process completion for gdb

[Sven Wischnowsky]

Clint Adams wrote:

> > I thought about something like that when I wrote _gdb.  The problem is
> > if there is no file matching *core.  Then the first would complete all
> > files (or directories), too.  And since they may be in different groups,
> > they would be shown twice.
> > 
> > Obviously, I didn't find a solution...
> 
> One could rely on the output of file(1), but that's ridiculously expensive.
> Why not an option to _path_files that doesn't add any matches if the
> pattern is not met?

That's what _path_files always does, only _files will re-try with
directories and all files.

_gdb is rather old and because we've added all that tag-order and
file-patterns stuff in the meantime, I would almost suggest to just use
the `_files -g "*core"' you suggested.  Users can then always use the
two styles to change that if they want (we could add a comment to _gdb
about that, asking users to tell us why they need it -- if the way their
systems handle core dumps is the reason for it).  Of course, they can
now use the file-patterns style to change it to use that pattern, too.

As I said, I had thought about these things when I wrote _gdb, I just
wasn't sure if `*core' is a good enough pattern for all systems.  And I
probably didn't ask back then.

So, I would be happy if you choose to change the call to _files, and
since noone complained that that would break core file completion on
their system...


Bye
  Sven


-- 
Sven Wischnowsky                         wischnow@informatik.hu-berlin.de

PATCH: gdb corefile completion [was Re: Bug#99095: Process completion for gdb]

[Clint Adams]

> So, I would be happy if you choose to change the call to _files, and
> since noone complained that that would break core file completion on
> their system...

Well, here it is.

Index: Completion/Unix/Command/_gdb
===================================================================
RCS file: /cvsroot/zsh/zsh/Completion/Unix/Command/_gdb,v
retrieving revision 1.1
diff -u -r1.1 _gdb
--- Completion/Unix/Command/_gdb	2001/04/02 11:51:19	1.1
+++ Completion/Unix/Command/_gdb	2001/06/01 12:11:27
@@ -47,7 +47,7 @@
   done
 
   if [[ $#w -gt 1 ]]; then
-    _alternative 'files:: _files' "processes:: _pids -m ${w[1]:t}"
+    _alternative "files:: _files -g '*core'" "processes:: _pids -m ${w[1]:t}"
   else
     _description files expl executable
     _files "$expl[@]" -g '*(-*)'