<div dir="ltr"><div class="gmail_extra"><div class="gmail_quote">On 19 March 2018 at 19:56, Chandler Carruth <span dir="ltr"><<a href="mailto:chandlerc@gmail.com" target="_blank">chandlerc@gmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div class="gmail_quote"><span class="gmail-"><div dir="ltr">On Thu, Mar 15, 2018 at 11:10 AM David Blaikie via llvm-dev <<a href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr">+Chandler who might have some thoughts on this.<br></div></blockquote><div><br></div></span><div>FWIW, I have no concerns about updating to a modern googletest. More modern the better IMO if someone is willing to do the work to make sure it works on all our platforms, etc.</div></div></div></blockquote><div><br></div><div>I'd be happy to do some of the work of this, and testing using MSVC14, perhaps 15 too, on x86_64 Windows 10 and (fairly elderly) GCC, possibly clang too, and Ubuntu 14.04, but I don't have easy access to other platforms/toolchains etc, so I'd need some assistance for this.<br></div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div class="gmail_quote"><div><br></div><div>However:</div><span class="gmail-"><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr">Could you provide an example here of the motivation for the feature you're missing?</div></blockquote><div><br></div></span><div>I definitely think that the whole space of combinatorial testing should be carefully discussed. Seems like a high risk of long-running tests w/o providing much incremental value.</div></div></div></blockquote><div><br></div><div>These are fair comments, and probably should be judged on a case-by-case basis.<br><br>In my particular case from <a href="https://reviews.llvm.org/D44560">https://reviews.llvm.org/D44560</a>, I currently test the following 3 different cases across the full set of DWARF versions and formats:<br></div><div> - Parsing a valid line table<br></div><div> - Emitting an error if the stated prologue length is greater than the actual length<br></div><div> - Emitting an error if the stated prologue length is shorter than the actual length<br><br></div><div>The first is just testing the positive cases for each possible input. I guess a single test for DWARF64 could be written for versions 2-4 and another for version 5 (where there is more stuff between the two length fields so this becomes interesting), similarly for DWARF32. To summarise, I think that these cases are interesting: V5 + 32, V5 + 64, V2 + 32/64, V3 + 32/64, V4 + 32/64. The biggest issue I have with cutting down to just this set is that it makes the tests less specific, e.g. at a glance, what is important about the test - the fact that it is v4, or DWARF64, or both independently, or the combination?<br><br>Related aside: I've realised since earlier that there is scope for version 2 tests, distinct from version 3: v2 tests test the lower boundary on valid versions, and v3 the upper boundary on versions without maximum_operations_per_instruction.<br><br>The latter two test cases are important because a) the length field has a different size for DWARF32/64 and therefore the prologue length needs to be measured from a different point between the different formats, and b) the contents of the prologue are different in each of version 3, 4, and 5, and thus the amount read will be different. We could test each individual version, independently of the format, but it is theoretically possible for an error to sneak in whereby the two different failure points cancel each other out. The benefit is admittedly small, but these tests are fast, so I don't think it hurts to have them.<br><br></div><div>James<br></div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div class="gmail_quote"><span class="gmail-"><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"> Might help motivate the discussion (and/or we'll end up nitpicking how it could be done differently without that feature... - which is sort of where I'm going with this. Combinatorial test case expansion does seem a bit suspicious to me - I'd hope we could pick a few examples from the various equivalence classes & that would suffice?)</div><div dir="ltr"><br><br><div class="gmail_quote"><div dir="ltr">On Mon, Mar 12, 2018 at 9:01 AM James Henderson via llvm-dev <<a href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div><div><div><div><div>Hi all,<br><br></div>I'm currently writing some unit tests for some debug line error handling code I'm working on (see e.g. <a href="https://reviews.llvm.org/D44382" target="_blank">https://reviews.llvm.org/<wbr>D44382</a>), and I just ran into an annoying disabled feature in gtest, specifically the "Combine" feature for use in combinatorially generating parameterised tests. A FIXME comment in ProfileData\<wbr>CoverageMappingTest.cpp suggests that I'm not the only one to have tried and discovered that they cannot use this feature. The problem is that the version of googletest (v 1.8.0, released in Aug 2016) in the LLVM tree requires TR1 tuple support for this feature, which is not really supported in recent compilers, and has been explicitly disabled in our googletest CMakeLists.txt, thus disabling "Combine".<br><br></div>I did a bit of looking around, and v 1.8.0 is indeed the last officially tagged release of googletest. However, there has been a lot of development on the framework since that point, including a fix to enable use of Combine with std::tuple-supporting compilers. There have been a number of issue raised on the googletest issue tracker (see e.g. <a href="https://github.com/google/googletest/issues/1467" target="_blank">https://github.com/google/<wbr>googletest/issues/1467</a> or <a href="https://github.com/google/googletest/issues/1079" target="_blank">https://github.com/google/<wbr>googletest/issues/1079</a>) asking about a 1.9.0 release, and there has been zero response from anybody answering the query of when/if it will happen. In the meantime, the last release gets older and crustier...<br><br></div>I'd therefore like to propose something that might be seen as slightly controversial: update to use ToT googletest (or at least some reasonably recent version of master), at least until a new release is created.<br><br></div>Thoughts?<br><br></div>James<br></div>
______________________________<wbr>_________________<br>
LLVM Developers mailing list<br>
<a href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a><br>
<a href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev" rel="noreferrer" target="_blank">http://lists.llvm.org/cgi-bin/<wbr>mailman/listinfo/llvm-dev</a><br>
</blockquote></div></div>
______________________________<wbr>_________________<br>
LLVM Developers mailing list<br>
<a href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a><br>
<a href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev" rel="noreferrer" target="_blank">http://lists.llvm.org/cgi-bin/<wbr>mailman/listinfo/llvm-dev</a><br>
</blockquote></span></div></div>
</blockquote></div><br></div></div>