On 2/10/2012 10:14 AM, Lionel Dricot wrote:
1) Indexing consume too much memory, making it impossible to use on less
than high-end server.
If a project is going to be large enough that indexing memory is a
serious issue, then I suspect that people already have beefy machines
for building it. I'm not saying it's a useful goal; I'm just saying that
it probably won't drive off developers.
2) DXR is too tight to mozilla-central
As a rule of thumb, if it's something specific to Mozilla, it should be
able to be placed in a customizable "plugin" of some kind.
3) Indexing is currently too slow (is it really a problem?)
4) DXR is hard to install and to use.
5) I still don't have a clear picture on how to use one DXR instance with
multiple repositories
This, IMHO, is the largest issue I would like to see fixed. At the very
least, the ability to have multiple trees in a single $DXR_ROOT install
and have it work nicely is necessary; even more feature-ific would be to
be able to share some index data cross-refs between different trees.
For 4), Carlos think that DXR should use autotools (or, I would say,
python-distutils). That would allow easy installation and, even more,
distribution packages. We should also make some clear shell commands :
index, deploy. Even a small PyGTK gui might be done quickly. Anyway, lot
of opportunity there.
The main problem that keeps an indexer from being easily packageable is
the fact that you have to do custom steps to compile the target source
code anyways; this is what would make producing a, say, Debian dxr
package difficult. Another option could be to do something like what
bugzilla does: give a web-configurable administration page (which can be
locked down, of course) to avoid ever having to physically see the
command line.
So, what are, in your opinion, the priorities of DXR as an OpenSource
project? Should we try to make a roadmap?
Another issue I recall from when I discussed this at the LLVM
developer's meeting was the "#ifdef issue": how do you handle the fact
that, after preprocessing, not all the original source code gets "seen"
by the compiler? The mental model I've had for a long time is to somehow
be able to merge the records of two different compile runs into a single
indexing output.
_______________________________________________
dev-static-analysis mailing list
[email protected]
https://lists.mozilla.org/listinfo/dev-static-analysis
