[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
RE: [aspectj-dev] ajdoc update

Hi George,

 

1: I have only been testing on Windows/MSIE (and I’m pretty sure that’s the case for Ron as well).  Could you create a bug report for the Opera/Mozilla/UNIX problem and add that patch to it?  Glad you caught that.

 

2: As described below this is the enhancement request 57974.  I just got your patch for it, thanks.

 

I can review and add these patches tomorrow after 1pm PST.  Before I do that, have you checked that all of the generated documentation in “ajdoc/testdata/coverage”, and “…/patterns”, and “.../spacewar” work peroperly?  Recall that I couldn’t apply your patch from Thursday (8th) because that broke the links in the “…/coverage” test case and “atrack”.  If all works for you let me know, and I can test against “atrack” if you don’t have it set up.

 

Mik

 


From: aspectj-dev-admin@xxxxxxxxxxx [mailto:aspectj-dev-admin@xxxxxxxxxxx] On Behalf Of George Harley1
Sent: Monday, April 12, 2004 1:39 PM
To: aspectj-dev@xxxxxxxxxxx
Subject: RE: [aspectj-dev] ajdoc update

 


> George: your suggestion of relativizing links is a good one.  However, the fix for that not only needs to check what package path it’s relativizing,
> but also where the type it is linking from is in the folder hierarchy (since the HREF base is taken as the current file you’re in, not the file that you’re pointing to).


Yes, indeed. Have another look at the patch I sent last Thursday (8th) : the relativizing is being done based on the package path of the "linking from" type.  


> But the key thing for this release is that the links work.


A couple of points :-

1. For me links do *not* work on Windows using the Mozilla or Opera browsers. I would also hazard that links will be broken on the Unix platforms as well with Mozilla and Opera. What browsers/platforms did you use to test that the links work ?

2. The current version of the link-setting code in CVS HEAD uses the java.io.File method getAbsolutePath() to set the "advises:" and "advised by:" links. As well as breaking the links for non-IE browsers (see point 1 above) this means that the resulting HTML will be broken the moment it is zipped up and moved around to another destination (e.g. like the AspectJ API Javadoc is).  

I will resubmit last Thursday's patch to you off-list (I think everyone else must be sick to death of this discussion by now). It will be in a slightly different form as the content of CVS HEAD has changed since then.

Hope this helps,
George
________________________________________
George C. Harley



"Mik Kersten" <beatmik@xxxxxxxxx>
Sent by: aspectj-dev-admin@xxxxxxxxxxx

09/04/2004 01:06

Please respond to
aspectj-dev

To

<aspectj-dev@xxxxxxxxxxx>

cc

 

Subject

RE: [aspectj-dev] ajdoc update

 

 

 




I’ve added Jan’s patterns to ajdoc’s test suite, and generating documentation for those works now.  I didn’t do a close inspection, but it looks right and links work.  The main fix needed to get that going was that ajdoc depended on packages names not ending with “.java”, which the pattern packages do.  Note that the test case uses the patterns out of docs/sandbox/ubc-design-patterns, which is kinda odd, but probably better than duplicating all those dozens of source files in CVS.
I fixed the following bugs
57780: ajdoc package summary should not show synthetic information
57773: ajdoc crash
 
Ron: all of the atrack documentation and links work for me now, and all your reports should be fixed, could you verify?  Thanks again for the thorough testing—it has made a huge difference for getting the quality up before others try it.
 
George: your suggestion of relativizing links is a good one.  However, the fix for that not only needs to check what package path it’s relativizing, but also where the type it is linking from is in the folder hierarchy (since the HREF base is taken as the current file you’re in, not the file that you’re pointing to).  I’m out of time on ajdoc (have spent 3 full days on it since the weekend--that’s 2 more than I had ;).  But the key thing for this release is that the links work.   The original ajdoc didn’t relativize, and so I’ve added that as an enhancement request (57974).  If you decide to do it, make sure that the testdata/coverage/doc documentation generated by running CoverageTestCase all works—it has the various variations of package/file/advice.  Btw, I cleaned up the link generating code so that’s much more simple.  I also addressed the other part of the problem that you were seeing—the code generation was not producing deeper than one level of type nesting properly, which is fixed and added to the coverage now.
 
Mik

 



From: aspectj-dev-admin@xxxxxxxxxxx [mailto:aspectj-dev-admin@xxxxxxxxxxx] On Behalf Of Mik Kersten
Sent:
Wednesday, April 07, 2004 10:27 PM
To:
aspectj-dev@xxxxxxxxxxx
Subject:
RE: [aspectj-dev] ajdoc update

 
Ron: thanks for all the great bug reports.  I should have time to get through most of them first thing tomorrow.  I downloaded atrack and am testing against it.
George, Andy: the relativization stuff is a good idea, but the latest patch is broken too.  The problem with relativizing this way is that the links are dependent on how deep you are in the folder hierarchy.  I’m going to fix that tomorrow, and expand the CoverageTestCase to include the various permutations of navigation that need to work.  If you find that after my fixes some example you have doesn’t work please extend that test’s testdata.  From this point on any changes to link generation should be manually checked (i.e. every link in CoverageTestCase’s generated documentation should work).  The manual checking is lame, but automating will take some time.  Running a link checker on it would be a good first pass…
Mik

 



From: aspectj-dev-admin@xxxxxxxxxxx [mailto:aspectj-dev-admin@xxxxxxxxxxx] On Behalf Of Ron Bodkin
Sent:
Wednesday, April 07, 2004 10:09 AM
To:
aspectj-dev@xxxxxxxxxxx
Subject:
RE: [aspectj-dev] ajdoc update

 
ajdoc is a lot closer to building aTrack now (thanks Mik!). I think I'm now running into the same issue George ran into with inner types, except that it causes a stack trace (probably because this is a doubly inner type!). Per Mik's request I've submitted a bugzilla report: https://bugs.eclipse.org/bugs/show_bug.cgi?id=57773
 
Ron
 
p.s. I'd suggest adding ajdoc and ajbrowser as components (as well as a development version) to Bugzilla.
 
 
------------Original Message------------
From: George Harley1 <GHARLEY@xxxxxxxxxx>
To: aspectj-dev@xxxxxxxxxxx
Date: Wed, Apr-7-2004 9:20 AM
Subject: RE: [aspectj-dev] ajdoc update

Hi Mik,

> I had to roll back some of Andy's fix to broken links:
> getRelativeComponent didn't work properly for internal anchor links (..html#<mumble>).


When I did a re-spin earlier today I found that ajdoc was producing HTML with broken links between advice in package A and advised members of types in package B.

Links between advice and advised types in the same package were OK.

The links to inner types were broken as well and the displayed text string for the links were also incorrect (e.g, if linking to packagefoo.ParentType.InnerType.membermethod() the visible text said it was linking to packagefoo.InnerType.membermethod()).

I sent a patch to Andy earlier. Let's see how it fares in the build.

Bye for now,
George
________________________________________
George C. Harley

"Mik Kersten" <beatmik@xxxxxxxxx>
Sent by: aspectj-dev-admin@xxxxxxxxxxx

07/04/2004 16:31

Please respond to
aspectj-dev

 

To

<aspectj-dev@xxxxxxxxxxx>

cc

 

Subject

RE: [aspectj-dev] ajdoc update


 

 

 






Based on Ron's feedback from running ajdoc on aTrack I've made the following
enhancements and a bug fix for dealing with declare forms.  Ron: please
Bugzilla so that this stuff gets recorded.  All: use the new -XajdocDebug
flag to send offending generated source files along with bug reports.

- Added support for 1.4 sources (use "-source 1.4" option)

- Made output class files go to documentation directory (-d) rather than
source directory.

- Added "-XajdocDebug" option that will keep the source files generated from
the ASM around in order to help people send better bug reports.

- Any declare forms that ajdoc doesn't know how to generate docs for get
printed to the console and indicated as a limitation.

I also extended test coverage to declare forms and added proper clearing of
state for multiple invocations from test cases.

Mik

> -----Original Message-----
> From: aspectj-dev-admin@xxxxxxxxxxx [mailto:aspectj-dev-admin@xxxxxxxxxxx]
> On Behalf Of Mik Kersten
> Sent: Tuesday, April 06, 2004 9:20 PM
> To: aspectj-dev@xxxxxxxxxxx
> Subject: [aspectj-dev] ajdoc update
>
> I just committed the updates to ajdoc needed for 1.2.  In a nutshell it's
> moved forward a ways from the previous ajdoc:
> - AJDT integration works
> - generates documentation for all the AspectJ modules
> - works on 1.3 and 1.4
> - comments and Javadoc tags are properly preserved and resolved
>
> Note that the hope is still for "declare" documentation to come in the
> form
> of a contribution.
>
> Getting it to work for our tree involved a bunch of bug fixes to deal with
> things like anonymous types.  To run from our tree in bootstrap mode use
> something like the attached batch script.  To run from AJDT point Eclipse
> at
> this command under the "Project -> Generate Javadoc" menu.  End users will
> be able to point at the "<aj-install-dir>/bin/ajdoc.bat".
>
> I had to roll back some of Andy's fix to broken links:
> getRelativeComponent
> didn't work properly for internal anchor links (..html#<mumble>).  Andy,
> could you check this against the current thing against whatever was
> breaking
> for you, and ideally add that code patter to CoverageTestCase?
>
> I added -XjavadocsInModel to make the addition of Javadoc strings to the
> ASM
> only happen when request it (prevent bloat of the model with strings.  The
> running time of ajdoc is now dominated by the compilation time.
>
> I had to make a change to a test case that I'm not quite comfortable with.
> AsmBuilderTest.testNullHandlingOfVisit() now checks for an NPE instead of
> an
> EmptyStackException.  I'm not sure how the original could have worked in
> the
> absence of the handle cache.
>
> Mik
>
> --
> http://kerstens.org/mik


_______________________________________________
aspectj-dev mailing list
aspectj-dev@xxxxxxxxxxx
http://dev.eclipse.org/mailman/listinfo/aspectj-dev