gotests
gotests copied to clipboard
cweill
generate benchmark for functions
Coverage decreased (-0.2%) to 95.44% when pulling 7136015a179955ba303920ddb0a7c4dde6707512 on mindlesslcc:master into 97d0977c0970fc63c7b864f3b50f2e37083dbf66 on cweill:master.
Hi @mindlesslcc, this is an impressive PR. You'll have to give me until this weekend to review this, as I'm swamped this week.
In the meantime, perhaps you can help me answer a more philosophical question: should generating benchmarks be a feature in a binary designed for generating test code?
Actually,benchmark is part of test, and sometimes I want to write some tests and benchmarks,and I think should give user a ability to generate benchmarks. For example, user can use a -b flag to specify generate benchmarks and It will not generate benchmarks default.
@mindlesslcc I stumbled across this project randomly but I spent some time looking at this PR.
I suppose this is ironic considering that this project is about generating tests, but it seems that many of tests added in this PR are redundant. I spot-checked about 4 of the golden test files by hand and they are testing nearly identical output with s/t/b/ and s/test/benchmark/ between tests and benchmarks.
I would speculate these could be tested against the original goldens by reading them into a buffer and doing a string replacement (or perhaps they could be re-generated using replacement tokens), and then you would not need two copies of each file, and test output would only need to be updated in one place.
The same appears to be true for internal/render/templates/function.tmpl and internal/render/templates/benchmark.tmpl, which are nearly the same. They could likely be squashed into a single template including variable names (b, bb) and output type (test or benchmark) for simplicity.
If the two templates are merged, I believe the the majority of these tests would be obviated since they are simply supplying different variables to the same code path. The variable functionality could be verified once and then the 38 or so permutations can be skipped.
The size of the PR would shrink to maybe 10 files with straightforward changes and @cweill would have a much easier time reviewing it. Thoughts?
