Sure, as long as we aren't turning the output of the "break set" or other 
commands into test API - which your version of this lit test does not - then 
this sort of test seems fine to me.

When we were starting out and building up the set of tests available I at least 
tended to err on the side of testing the whole lifecycle of the subsystem I was 
testing.  That way you could get a lot of coverage per test, which we really 
needed.  So there probably are more tests that run end-to-end than absolutely 
necessary.  We still should make sure, particularly when we are adding new 
functionality, that there are enough end-to-end tests before we start testing 
the peripheries of the feature.  But for things like breakpoint setting, we are 
well past that point, and writing lots of setting only tests is appropriate, as 
you say.

BTW, one thing I like about writing tests is that it is easy to craft 
fairly rich failure messages so if you get errors on systems you don't have 
access to or are dealing with something that fails intermittently on a bot 
somewhere, you have a hope of figuring out what went wrong.  Is this possible 
with FileCheck tests?


> On Feb 23, 2018, at 1:52 PM, Pavel Labath <> wrote:
> On 23 February 2018 at 11:24, Jim Ingham <> wrote:
>> To be fair, you could probably have made the version of the test 
>> close to as fast by not running a process.  The old test was testing that we 
>> got a location for the breakpoint AND hit it.  The latter was probably 
>> overkill.  But the two tests aren't testing the same thing.
> Yes, they are not testing the same thing. However, my thought was (and
> it sounds like you agree) that for checking case-sensitivity checking
> we are able to resolve the breakpoint to a file address is enough. We
> have plenty of tests that verify the path from a file address to an
> actual breakpoint hit.
> The reason I went this route (instead of just modifying the original
> test) is that this will make it easy to write more low level
> breakpoint tests. For example, you can write an assembly file with
> amusing .loc directives and check that lldb resolves that reasonably
> (this looks like it could be useful for the ppc patch). Adding llvm-mc
> support to would be possible, but more difficult than with a
> test like this.

lldb-commits mailing list

Reply via email to