labath added a subscriber: jankratochvil. labath added a comment. In https://reviews.llvm.org/D48782#1149051, @plotfi wrote:
> In https://reviews.llvm.org/D48782#1148498, @alexshap wrote: > > > @labath > > > > > I am not denying that there is value in running the dotest suite in all > > > of these modes. In fact, I think that (the fact that we can use the same > > > tests to exercise a lot of different scenarios) is one of the strengths > > > ?>of our test suite. However, I don't believe all of these modes should > > > be inflicted onto everyone running lldb tests or be our first and only > > > line of defense against regressions. > > > > for what it's worth - not sure how much you care about my opinion, but i > > think it's an important point but it doesn't actually contradict or prevent > > your second point regarding adding regression tests using lldb-test, > > however i think those should be added over time (sadly no tests were added > > when the support for .dwp was implemented / introduced) (not in this patch). > > I think that the approach of this patch is still useful, this mode can be > > off by default, but if smb needs to run all the tests with dwps - it's easy > > to do by passing or setting a variable (for example). > > > yes, thats the near term solution. What's the medium-to-long term solution? > In https://reviews.llvm.org/D48782#1148929, @JDevlieghere wrote: > >> In https://reviews.llvm.org/D48782#1148376, @labath wrote: >> >> > Then, for the integration test part, I propose to come up with a more >> > generic way to specify the kind of debug info to generate. I don't have >> > this fully thought out yet, but I have been thinking of something that >> > could wrap the compiler invocation with some user specified script. Then >> > we could use the same mechanism to run DWP and DWZ with any kind of crazy >> > compiler flags we think of (which is definitely useful when bringing up >> > support for a new kind of debug info format). If someone has a particular >> > interest in a specific combo, he can set up a bot which runs tests in this >> > mode (details about who would be responsible for fixing failures TBD) >> >> >> I really like that idea. It sounds similar to the EXPENSIVE_CHECKS we have >> for LLVM. Personally I'd love to have an overnight job that runs the test >> suite with DWARF5 for example. > > > Oh, I see. Have some generic option for debug format to make this problem > simpler every time it arises. The thing is, we *already* have multiple needs for that. Maybe you could work with @jankratochvil to come up with a solution that would work for both DWP and DWZ? It sounds to me like both of these formats (and even dSYM) could be described by "there is some tool which runs after we link the main executable/library". If we could have a generic way to specify this as an argument to test suite, then we could solve both problems together. As the tools are unlikely to take the same arguments (and IIRC, for DWZ there are multiple commands we need to issue), we can have wrapper scripts (checked into the utils folder or somewhere), which hides these differences. Does that make sense? https://reviews.llvm.org/D48782 _______________________________________________ lldb-commits mailing list lldb-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/lldb-commits
