[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
[
List Home]
|
Re: [atf-dev] Breakpoint matching in the ATF debugger
|
Giuliano Mega wrote:
Greetings,
My name is Giuliano Mega, and I am currently doing some work on the
ATF debugger. :-) I am a committer at the Google Summer of Code Project,
where I've implemented an Eclipse debugger for distributed object
applications (http://god.incubadora.fapesp.br/portal).
My current objective with the ATF debugger is to get eval support
working well (already checked out the eval_support branch), and to
that end I have been working on getting some automated tests in place
(including a failing test case for eval).
During the course of that work, I realized that
JSDebugThread#getBreakpoints is not functional, so I am trying to
implement it. I am having some trouble, however, with matching the
information that the Mozilla engine provides with what's available from
the IBreakpoint instances.
It seems that it is not possible to ask the engine to assign a unique ID
to each breakpoint, so the only way to get this working, as far as I can
tell, is trying to match the path structure of the URL returned from
jsdIStackFrame#getScript#getFileName against the path structure of the
resource returned from IBreakpoint#getMarker#getResource. Right? :-)
Right, except when it's not ;-). The Mozilla engine labels eval buffers
with the
file name of the function that calls eval(). These line numbers overlap
the original file.
So for breakpoints in eval() buffers you can't use this approach.
What I did in Firebug 1.2 was create a covering object for each
compilation unit.
(eg file, eval-buffer, event-buffer). Each unit has a unique id, a
URL-like string. For
files, the ID is the jsdIScript.fileName; for eval buffers its the
jsdIScript.fileName plus
"/" plus the MD5 hash of the source in the buffer. Just after a unit is
compiled, I look thru the
list of breakpoints for any that apply to this unit. If yes, I set the
breakpoint on the jsdIScript.
If you want to look at the Firebug source, you can probably make sense
of it by working
back from the mozilla APIs calls shared with your project. If you do,
for sure use
http://fbug.googlecode.com/svn/branches/firebug1.2
as this code makes much more sense than the 1.1 version. At least I
hope ;-)
Unfortunately, the structure of the URL doesn't exactly mirror the path
structure of the resources in the workspace. For instance, when I get an
URL "http://localhost:8080/DebugTests/test.js";, it refers to file
"workspace/DebugTests/WebContent/test.js".
So, I was wondering. Is it safe to assume that this transformation is
regular, I mean, that it's always performed in the same way for all
servers we may want to debug on? If so, then the problem is easy to
solve, and my current implementation will work fine. If not, then is
there a standard way to know which kinds of path manipulations has the
deployment process performed?
I created my own solution for this in fireclipse; any solution will be
http-server dependent,
but eclipse could and should offer a standard mapping for all
eclipse-embedded or
configured servers. You may need to ask on one of the Wxx eclipse forums.
John
--
______________________________________________________
John J. Barton email: johnjbarton<at>johnjbarton.com
http://www.johnjbarton.com
