[lldb-dev] [cfe-dev] ASTImporter patches and improvements, please help
Gábor Márton via lldb-dev
lldb-dev at lists.llvm.org
Thu Dec 6 06:39:08 PST 2018
Hi Davide,
Thank you for your email.
> In particular, what's the error you get when lldb fails immediately
running the tests?
> Also, have you checked libcxx and libcxx-abi in your build? We might
> consider making that a mandatory dependency for the Cmake build.
Finally I could run the test suite ("ninja check-lldb) on our macOS without
errors. Here are my findings:
1) "sudo /usr/sbin/DevToolsSecurity --enable" is a must.
2) I indeed did not checked out libcxx. (libcxx-abi was not needed to have
a successful run). Would be great to make that a mandatory dependency.
3) Besides that I recognized that "ninja check-lldb" target does not build
all it's dependencies. E.g. lldb-server tests failed until I executed
"ninja debugserver" before "ninja check-lldb". I am not sure, but I think
"ninja obj2yaml" and "ninja llvm-pdbutil" are also prerequisites (and these
targets seem not to be built by a simple "ninja" command).
Thanks,
Gabor
On Sat, Dec 1, 2018 at 9:40 PM Davide Italiano <dccitaliano at gmail.com>
wrote:
> On Sat, Dec 1, 2018 at 8:00 AM Gábor Márton via cfe-dev
> <cfe-dev at lists.llvm.org> wrote:
> >
> > Dear LLDB Developers,
> >
> > There is an ongoing activity to improve ASTImporter in Clang to make it
> support
> > C++ (and C) completely and correctly. Our team works on cross
> translation unit
> > (CTU) static analysis where we use the in-tree Clang static analyzer in
> > combination with the ASTImporter (via the scan-build.py script or
> CodeChecker).
> > In the past 18 months we have committed more than 70 patches to make the
> > ASTImporter better. Our primary target is Linux and we do run the LLDB
> tests
> > on Linux before we commit. Unfortunately, quite recently 3 of these
> patches
> > broke the LLDB macOS buildbots (green.lab.llvm.org/green/view/LLDB) and
> this
> > caused some turmoil. We are very sorry about this and we are making
> actions to
> > avoid such inconveniences in the future: We ordered a Mac Mini, until it
> > arrives we are using a borrowed Mac Book. We are going to create a CI
> job which
> > will execute the macOS LLDB test suite for a specific patch. Besides
> this, for
> > every patch we are going to monitor the macOS buildbots once they are
> > committed.
> >
> > However, we are experiencing problems both with the buildbots and with
> the LLDB
> > tests, therefore we are asking help from the LLDB community in the
> following:
> >
> > (1) Apparently, the green lab macOS buildbots are not displayed in the
> build
> > bot console (http://lab.llvm.org:8011/console). More importantly they
> are not
> > capable of sending emails to the authors of the commit in case of
> failure.
> > Would it be possible to setup the these buildbots to send out the emails
> for
> > the authors?
> >
>
> Thanks for reporting this problem. I cc:ed Mike, as he owns the bots,
> and he should be able to help us with this.
> (I do have access to the Jenkins configuration but I don't feel
> confident enough to make the change myself).
>
> > (2) We are facing difficulties with the LLDB lit tests both on macOS and
> on
> > Linux. E.g. on a fresh macOS mojave install, checking out master LLVM,
> Clang,
> > LLDB and then executing ninja check-lldb fails immediately. On Linux we
> > experienced that some tests fail because they try to read a non-existent
> > register. Some tests fail non-deterministically because they can't kill a
> > process or a timeout is not long enough. Some tests fail because of a
> linker
> > error of libcxx. We would like to address all these issues. Would you
> like to
> > discuss these issues on lldb-dev or should we create a bugzilla ticket
> for
> > each?
> >
>
> I think either of them is fine. If you're willing to spend some time
> on this, I'm sure people on the LLDB side will consider helping you.
> In particular, what's the error you get when lldb fails immediately
> running the tests?
> Also, have you checked libcxx and libcxx-abi in your build? We might
> consider making that a mandatory dependency for the Cmake build.
>
> > (3) ASTImporter unit tests and lit tests in Clang for the LLDB specific
> usage
> > are very-very limited. LLDB uses the so called "minimal" import, but
> none of
> > the unit tests exercises that (CTU uses the full import). We should
> have a
> > unit test suite where the LLDB use cases are covered, e.g. after a
> minimal
> > import an `importDefinition` is called. Also, a test double which
> implements
> > the ExternalASTSource could be used. I believe it would be possible to
> cover
> > with the unit tests all LLDB/macOS related scenarios and these tests
> could run
> > on Linux too. We do something similar in case of windows: we execute
> all unit
> > tests with "-fdelayed-template-parsing" and with "-fms-compatibility"
> too. I
> > think, the more unit tests we have in Clang the sooner we catch the LLDB
> > specific importer errors. I am asking the LLDB community's help here,
> because
> > we lack the domain knowledge in LLDB so we don't know what expectations
> should
> > we write in each unit test cases. (About lit tests, there is
> clang-import-test
> > but its coverage seems pretty low and I don't know if that really
> exercises all
> > LLDB use cases.)
> >
>
> Definitely. +1.
>
> --
> Davide
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/lldb-dev/attachments/20181206/e37e401e/attachment.html>
More information about the lldb-dev
mailing list