CEDET

>>> yary <not.com@...> seems to think that:
>I just finished a smallish c-project where I used a little bit of
>CEDET, and a fair amount of etags & find-tags M-.
>
>Now I'm working on a small-medium python project, and would like to
>use more of Semantics smarter tag handling.
>
>How do I generate tags/mass bovinate for all source files in a
>directory & subdirectories, and use an equivalent of find-tags
>(presumably in a CEDET tool) that knows where to look better than
>etags?
>
>I've seen this question before but can't find it in my archives, or on the web.
  [ ... ]

Hi,

  There is a script, semanticdb.sh, which you can use to pre-parse
lots of files.  If you are starting the project from scratch, then
there is no need to pre-parse anything if you always edit in Emacs.  I
haven't used this in a while (ie, years), so am not sure of it's
current state.

  If you use the CVS version of CEDET, then a change I made over the
weekend will do the mass parsing of files for tags in idle time.
Thus, if you are editing a file, then get up to get a cup of coffee,
it will have then pre-parsed all files relevant to what you were
working on.

  I'm not too sure about the python part though.  The C code support
knows about C style include paths.  I (personally) know nothing about
python, and suspect you may need to add include paths.  You could do
this:

(defvar-mode-local python-mode semantic-dependency-include-path
  '("/some/path" "/to/code")
  "System path to search for include files.")

  where you point the path at your python library files, headers,
whatever python uses.  This will give you completion on system
libraries in Python.

  Once you have that set up, look in the Senator menu "Navigate".
There are two jump functions.  If there is more than one hit for a
name, press TAB, to show where each is so you can choose.

  Since I don't use Python, you may find it doesn't have a complete
set of configuration for Semantic.  Adding the right features won't be
too hard for someone who knows Python.

Eric

-- 
          Eric Ludlam:                       eric@...
   Siege: http://www.siege-engine.com          Emacs: http://cedet.sourceforge.net

I have tried the python support in cedet, and I think that it requires
more work. At mean time, you can try to install rope and ropemacs. I
have tried them myself and found that auto completion and refactoring
all work very well. The problem of ropemacs is that the interface can be
improved, but it is already mature. Ropemacs is written by python
with pymacs, so I guess it is much easier to support python using python
language. You can find ropemacs here.

http://rope.sourceforge.net/ropemacs.html

In my opinion, it would be the best if we can integrate rope with
cedet. 

Best,

Ming-Wei

"Eric M. Ludlam" <eric@...> writes:

>>>> yary <not.com@...> seems to think that:
>>I just finished a smallish c-project where I used a little bit of
>>CEDET, and a fair amount of etags & find-tags M-.
>>
>>Now I'm working on a small-medium python project, and would like to
>>use more of Semantics smarter tag handling.
>>
>>How do I generate tags/mass bovinate for all source files in a
>>directory & subdirectories, and use an equivalent of find-tags
>>(presumably in a CEDET tool) that knows where to look better than
>>etags?
>>
>>I've seen this question before but can't find it in my archives, or on the web.
>   [ ... ]
>
> Hi,
>
>   There is a script, semanticdb.sh, which you can use to pre-parse
> lots of files.  If you are starting the project from scratch, then
> there is no need to pre-parse anything if you always edit in Emacs.  I
> haven't used this in a while (ie, years), so am not sure of it's
> current state.
>
>   If you use the CVS version of CEDET, then a change I made over the
> weekend will do the mass parsing of files for tags in idle time.
> Thus, if you are editing a file, then get up to get a cup of coffee,
> it will have then pre-parsed all files relevant to what you were
> working on.
>
>   I'm not too sure about the python part though.  The C code support
> knows about C style include paths.  I (personally) know nothing about
> python, and suspect you may need to add include paths.  You could do
> this:
>
> (defvar-mode-local python-mode semantic-dependency-include-path
>   '("/some/path" "/to/code")
>   "System path to search for include files.")
>
>   where you point the path at your python library files, headers,
> whatever python uses.  This will give you completion on system
> libraries in Python.
>
>   Once you have that set up, look in the Senator menu "Navigate".
> There are two jump functions.  If there is more than one hit for a
> name, press TAB, to show where each is so you can choose.
>
>   Since I don't use Python, you may find it doesn't have a complete
> set of configuration for Semantic.  Adding the right features won't be
> too hard for someone who knows Python.
>
> Eric

After some trial-and-error getting cygwin's bash & /usr/bin/find &
emacs to agree on pathnames, I got semanticdb.sh to parse all my *py
files in all the subdirectories. It's an existing project, and it's a
collection of scripts called by php pages (not my design...), so
importing the various files was the only way. There's no "main()", no
single point of entry that includes all relevant libraries. Had to
either open them one-by-one in emacs or use that shell script.

You may get more questions like that when cedet is distributed with
emacs, and other folks use it to digest their legacy codebases. Or
maybe not, cross that bridge if you come to it.

On Mon, May 12, 2008 at 6:06 PM, Ming-Wei Chang <mchang21@...> wrote:
>
> I have tried the python support in cedet, and I think that it requires
> more work.

Speedbar does a good enough job of finding classes and methods. For
example, it knows that "save2AltXML" is a method of "class VarLog" in
file "VarLogDef.py" - however it took me a while to discover I needed
C-c , J (shift-J) and not C-c , j to find a tag in any file- thought
the batch parse hadn't worked out.

The good thing is that semantic-complete-jump tells me that there are
four "save2AltXML" methods in two files, whereas find-tags wouldn't
tell me that. True, the autocomplete doesn't work much in this
project, and semantic can't tell me which save2AltXML method will be
called (not even sure if that's possible before runtime), but I'll
live with that for now.

> At mean time, you can try to install rope and ropemacs. I
> have tried them myself and found that auto completion and refactoring
> all work very well.

Rope looks pretty cool, but I'm still jumping between c, python, perl,
and php, and won't be refactoring the python code (though it could use
it...)  Going to try and learn the cedet tools better since they are
more widely applicable, will install rope & ropemacs if and when the
python coding gets deep.

thanks for the tips, all

-y

>>> yary <not.com@...> seems to think that:
  [ ... ]
>On Mon, May 12, 2008 at 6:06 PM, Ming-Wei Chang <mchang21@...> wrote:
>>
>> I have tried the python support in cedet, and I think that it requires
>> more work.
>
>Speedbar does a good enough job of finding classes and methods. For
>example, it knows that "save2AltXML" is a method of "class VarLog" in
>file "VarLogDef.py" - however it took me a while to discover I needed
>C-c , J (shift-J) and not C-c , j to find a tag in any file- thought
>the batch parse hadn't worked out.

Hmm.  That keeps coming up.  If anyone has a suggestion on how to bind
the keys for this, let me know.  Just because I pick something for a
keybinding doesn't mean it has to stay that way.

>The good thing is that semantic-complete-jump tells me that there are
>four "save2AltXML" methods in two files, whereas find-tags wouldn't
>tell me that. True, the autocomplete doesn't work much in this
>project, and semantic can't tell me which save2AltXML method will be
>called (not even sure if that's possible before runtime), but I'll
>live with that for now.

You have run into an interesting problem I don't have a good
user-facing story for.  The semantic-complete-jump routines will find
any symbol most anywhere, and if you press TAB, it will show you which
symbol you have selected if there is a choice.

In the semantic-ia.el file, there are routines that use the analyzer
to get more information.  In there is semantic-ia-fast-jump.  This
will analyze the current context, and find exactly which of the 4
methods it might be, and jump there.  This routine, however, does not
support typing in random names.

There is probably a happy middle ground somewhere, though I'm not sure
on how to get it coded up for a clear user interface.


I've tended to focus on getting the back end stuff parsing, storing,
and analyzing correctly.  Many of the user interfaces in Semantic
right now just expose some corner of the more difficult logic.  Making
a good user interface is time consuming to build and maintain, like
ECB.  There is no way I could manage the back ends, and something as
complex as ECB.  I would expect that the ultimate jump to a symbol
function could be similarly complex.  Studying user patterns, how to
get to different goal destinations in the least number of keystrokes,
what symbology to use in the completion prompt, and keep it language
independent and dependent at the same time.

The confusion between the functions in CEDET that do this is
understandable, and Emacs will need a single super-jump/complete
function since this is a common new user issue.


Eric

-- 
          Eric Ludlam:                       eric@...
   Siege: http://www.siege-engine.com          Emacs: http://cedet.sourceforge.net

"Eric M. Ludlam" <eric@...> writes:

>
>>Speedbar does a good enough job of finding classes and methods. For
>>example, it knows that "save2AltXML" is a method of "class VarLog" in
>>file "VarLogDef.py" - however it took me a while to discover I needed
>>C-c , J (shift-J) and not C-c , j to find a tag in any file- thought
>>the batch parse hadn't worked out.
>
> Hmm.  That keeps coming up.  If anyone has a suggestion on how to bind
> the keys for this, let me know.  Just because I pick something for a
> keybinding doesn't mean it has to stay that way.
>

Maybe just a a helpnmessage when c-j , j doesnt bring up any condidates?

The prompt could be like:

Local Symbol[c-j , J for all symbols]


-- 
Joakim Verona

"Eric M. Ludlam" <eric@...> writes:

> You have run into an interesting problem I don't have a good
> user-facing story for.  The semantic-complete-jump routines will find
> any symbol most anywhere, and if you press TAB, it will show you which
> symbol you have selected if there is a choice.

There are two issues. In "dynamic" languages, and even in static
languages, there will be times when an object type will only be known
at runtime, so the several choices are all valid, and
semantic-ia-fast-jump will likely figure that out. Tabbing through the
choices is a good start.

Ideally, in that case, the choice builder would recognize that it has
identical items. Then it would walk up the syntax tree to find where
they differ- ie "These two tags called "X" are members of "Circle" and
"Square" so show Cirlce/Square after them" or "getName" exists in two
classes named "File" in "AccessLog/File.py" and "ErrorLog/File.py" so
show the class, file, and path in the choice list.

As for when to call semantic-complete-jump vs semantic-ia-fast-jump -
that sounds akin to find-tags (M-.) vs search-tags (M-,) - I'm
thinking of rebinding those keys to the semantic versions.


On Tue, May 13, 2008 at 5:09 AM,  <joakim@...> wrote:
> "Eric M. Ludlam" <eric@...> writes:
...
>> Hmm.  That keeps coming up.  If anyone has a suggestion on how to bind
>> the keys for this, let me know.  Just because I pick something for a
>> keybinding doesn't mean it has to stay that way.
>>
>
> Maybe just a a helpnmessage when c-j , j doesnt bring up any condidates?
>
> The prompt could be like:
>
> Local Symbol[c-j , J for all symbols]

Help message is good. Also think that more-common uses should take
less keystrokes. If global symbol search is more used than local, than
use shift-J for local and unshifted-j for global.... which will
confuse all the current users.