[PATCH] D45700: Improve LLDB's handling of non-local minidumps

[Leonard Mosescu via Phabricator via llvm-commits]

lemo added subscribers: bgianfo, penryu.
lemo added a comment.

> It looks like nobody except me is worried about the
>  module-without-an-object-file situation, so I guess we can try this out and
>  see how it goes.

Sorry Pavel, I meant to respond to this: most of the code seems to
explicitly handle this case (module-without-object-file), I just had to fix
a couple of cases. It's possible that more fixes will be required, but the
intention seems to be to accommodate for missing object files so at least
in this area I didn't have to break new ground.

I also considered creating placeholder object files, but it proved a bit
more intrusive since there are numerous places where it's assumed that
object files map to a real file which can be read and written to. Maybe at
some point we'll need to reconsider this (placeholder object files) but for
the initial iteration the placeholder modules seems to be sufficient. The
only downside I noticed is mostly cosmetic, for example things like "target
modules dump objfile" may print empty lines for the missing object files.

The test you've added here has been failing on windows though. I've tried

> to fix this in r330314, but it meant modifying the module.file.fullpath
>  output expectations. I'm not sure where you're going with the minidump
>  support, but if you are bothered by module.file.fullpath containing a
>  forward slash, you may want to look into fixing the SBFileSpec behavior.

Thanks!


Repository:
  rL LLVM

https://reviews.llvm.org/D45700