Re: Current state of python.el in the Emacs trunk

[ken manheimer <ken.manheimer@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 7615 bytes --]

On Thu, Feb 24, 2011 at 1:13 AM, Christoph <cschol2112@googlemail.com>wrote:

> Dave Love <fx@gnu.org> writes:
>
> > I'm not sure what fixes you mean.  I've just maintained a separate
> > version.
>
> I was volunteering in case you were willing to let us incorporate any
> fixes and improvements you made to your separate mode into the trunk but
> couldnt or didnt want to do it yourself. The commentary of your file
> says that certain bugs have been fixed, iirc.
>
> > That stuff simply shouldn't be there, which is why it originally
> > wasn't.
>
> That's what I am currently working on cleaning up. This is mainly to
> provide a cleaner base for the integration of functionality from
> Fabian's new mode.
>
> > It appears all the world's Python, so it doesn't matter if it causes
> > global problems and more-or-less excludes similar things for other
> > interpreters as far as I remember.  (I implemented a gud-minor-mode, of
> > course, but it required changes to gud.el and I didn't keep up with
> > Emacs.  I think the general facility was less code than the
> > Python-specific version.)
>
> I am unclear as to how gud-minor-mode would be better, since I am not
> familiar with that mode at all. However, I remember reading in some old
> thread where somebody made the point that Python's way of debugging
> doesn't lend itself well to using a gud-minor-mode. I might have to go
> back and reread that thread again.
>
> Christoph


sorry it's about a month later, but i wasn't aware of these conversations
until this eve.  i expect that the point to which you're referring is mine,
stated in the message included below.  the upshot is that, in my
not-so-recent acquaintance with gud and gdb-oriented debugging, they're
oriented to be used as executives, running the program being debugged.  this
(if it is, in fact, the case) is extremely cumbersome and limited when it
comes to debugging very dynamic languages, where you may want to embark on
debugging in the middle of an already-executing run, dropping to an
interactive session when a traceback is encountered.  this is useful for
many programs, and crucial for long running servers (like zope, which was my
main focus when i developed pdbtrack for python-mode.el).

when i originally developed pdbtrack i looked at gud integration, and it was
horrifying.  i got 90% of what i needed by using a comint output filter
which noticed the python debugger's (pdb's) characteristic output and
fetched the corresponding files in another buffer, with the overlay-arrow on
the current line, and tracking execution as you interact with the python
debugger in whatever the shell is where it's running.  i did that with less
that 150 lines of elisp lines, including a lot of comments.  i depend on it
daily when debugging python within emacs, and i think many other emacs /
python users do, too.

i was spurred to speak up when dave suggested in the thread (as i excerpted
in the following message) that "it's not difficult to restructure GUD" to
provide the functionality, and therefore "and it's not clean to make an
add-on, which is why it's [pdbtrack, i believe] not in python.el".  it
sounded like, by suggesting providing the functionality with gud, dave was
missing and/or dismissing some central features which make pdbtrack useful.
 i can see making it easy to disable for the people who don't want to use
it, but making it unavailable for those who need it was a very frustrating
prospect.  however, pdbtrack does seem to be present in the versions of
python.el that come with emacs 23 on the various systems i use, so i'm
hopeful it's been adopted and will continue to be.

ken

---------- Forwarded message ----------
From: ken manheimer <ken.manheimer@gmail.com>
Date: Thu, Feb 19, 2009 at 3:21 PM
Subject: Re: replacing python.el
To: Dave Love <fx@gnu.org>
Cc: Glenn Morris <rgm@gnu.org>, "skip@pobox.com" <skip@pobox.com>, "
rms@gnu.org" <rms@gnu.org>, "python-mode@python.org" <python-mode@python.org>,
"XEmacs-Beta@xemacs.org" <XEmacs-Beta@xemacs.org>, "barry@python.org" <
barry@python.org>, "fbe2@comcast.net" <fbe2@comcast.net>, "
emacs-devel@gnu.org" <emacs-devel@gnu.org>


sorry to be late to this python.el / python-mode.el convergence
discussion.  i'm concerned that what dave is describing would not
preserve a crucial feature of pdbtrack, a feature that may be a reason
some python developers choose python-mode.el over python.el.  (it is
for me.)

On Sun, Feb 1, 2009 at 3:51 PM, Dave Love <fx@gnu.org> wrote:
>[...]
> Indeed, and I don't understand what other problem there is with it,
> other than maintenance.  Why does it need to be replaced with
> python-mode.el, even if that was properly assigned?
>
> The only other worthwhile feature I know of sort-of from python-mode.el
> is related to something called pdbtrack (?).  My commentary explains
> that part of the functionality already exists, and something more
> general than the rest should be a general feature in GUD.  (The
> Python-specifics are already there.)  It's not difficult to restructure
> GUD -- or wasn't when I hacked it originally -- and it's not clean to
> make an add-on, which is why it's not in python.el.  I know there isn't
> interest in abstractions like that, but I didn't want to preempt a
> possible change of opinion.

when i last looked at it, gud was a terrible fit for pdbtrack.  if
things haven't changed drastically,  i'm concerned that what you're
suggesting would sacrifice pdbtrack's dynamicism.

this is all the more worrisome since i spent some time porting
pdbtrack to python.el.  it's currently there in emacs 23.0.90, but i
see it's not in the emacs that comes w/recent ubuntu, emacs 22.2.

here's the scoop, as best i can describe it.

gud is oriented to being in control of debugging, in particular
launching the program being debugged, or injecting a connection to
the executive to start the debugging at an arbitrary point, or
post-mortem.

pdbtrack, on the other hand, simply works wherevever python's
debugger, pdb, does.  the pdbtrack code is responsible for detecting
pdb activity within an (any) emacs shell and presenting, in a
companion buffer,  the source file and line that pdb is reporting as
the current instruction.

pdbtrack provides functionality for python's debugger, pdb, much like
edebug does for emacs lisp, except that pdb can be triggered as a
statement within the subject program's code, and not just taking over
execution of the program (as gdb and edebug do), or run
onerror/post-mortem.  (pdb also provides those modes.)  this turns out
to be invaluable in general, and especially for long-running programs
like servers, where you want debugging to trigger in very specific
situations.  with pdb, you just situate the debug-triggering code
exactly in the situation (i wish i could do this with edebug), or as
error handling around the situation.

gud may have facilities that could be used to enhance pdbtrack, but i
don't think it is designed to operate the way i describe above.  plus,
gud is massive, and i suspect it would take more emacs lisp code to
craft gud python accommodations than all of pdbtrack (last i checked
around 150 lines, including copious comments).

i hope this is clear.  as i said, pdbtrack porting to python.el has
already been done, it just isn't in the currently released version.
and, of course, there are other benefits to be had from a
python-mode.el/python.el convergence, but i want to make sure this
one, at least, is preserved.  (i also think the pdbtrack approach
would be appreciated for other dynamic languages, including emacs
lisp.)
--
ken
http://myriadicity.net


>

[-- Attachment #2: Type: text/html, Size: 10669 bytes --]