[clangd-dev] Replacing -lit-test with a test utility?

[Dan Liew via clangd-dev]

Hi,

> On 14 Aug 2018, at 16:16, Ilya Biryukov <ibiryukov at google.com> wrote:
> 
> Hi Jan,
> 
> I agree that '-input-style=delimited' and '-lit-test' does look like something that shouldn't be in clangd from the architectural standpoint, as it's only specific to the test.
> In practice, the python wrapper will probably be more complicated than what we have now and could even make some things harder, e.g. it's very easy to debug failures now and I could imagine problems with debugging a python wrapper.
> 

Not really sure about harder. If you avoid pipes (by writing the output/input to temp files using lit's shell-test `%t` substitution) and use files instead tests should still be easy top debug.

Apart from the architecture not being very clean with `-lit-test` the biggest problem is that if your tests always use `-lit-test` it means you're never actually testing `clangd` in the configuration
it's actually going to be used in during deployment. To me that seems completely unacceptable if you actually want to have confidence that the code that is deployed actually works.