On Tue, May 22, 2012 at 08:48:22PM +0200, Yann Dirson wrote:
> latrace is currently cool for simple programs that use just libc
or
> some lib small enough to make (re)writing a header file reasonable. I
> was looking forward to using it to investigate the interactions
> between a Qt program and the Qt libs - you will understand I don't
> intend to write (C-style) headers for (even a usable subset of) Qt :)
yep, it's allways a pain.. probably those libs are even worse ;)
Since I have not found a better way yet, I have started work on adding
some Qt support. My current plan looks like:
* managing to use a new DTYPE (not so easy, the STRUCT vs POD split is
hardcoded in many places)
* hacking some QObject support so we can just specify "QObject*" in
the headers and call into Qt introspection to get the real class shown
* adding specific support for useful QObject subclasses (eg. QString)
* adding C++ mangling support so we don't need to write mangled headers
* see if Qt metaobject provides enough info about methods to produce headers
on-the-fly
* offload DTYPE_CUSTOM processing to dlopenned modules
> Are there any plans to access those debug symbols (through
libbfd
> maybe ?) in latrace ?
sadly I haven't got much time to spend on latrace lately,
but I'll keep this in mind
we should get more developers attracted to latrace somehow ;)
After having worked through some parts of the code, I shall say that
some parts could probably attract more devs with a couple of comments :}
If you could for example explain what classificate() is expected to
do, what those ass_* thingies are, and the like, I'd be really grateful.
On another issue I have not dug into yet: C++ demangling appears not
to work at all on Debian, not sure if it is related to libiberty
version or what. Luckily c++filt is there :)