[PATCH] D113268: [clangd] Fix tests with forward slash as separator on windows

[Sam McCall via Phabricator via cfe-commits]

sammccall added a comment.

In D113268#3111995 <https://reviews.llvm.org/D113268#3111995>, @mstorsjo wrote:

>> I think the biggest testing problem we have though is that we don't have a good way of defining integration tests with paths set up the right way (since we can't really use absolute paths in our tests, and most of our bugs involve comparing absolute paths).
>
> Yup, I can totally understand that. And as I haven't tested clangd in this setup other than the unit tests, I'm not sure if it works entirely in practice though - with other unit tests in Clang I've seen lots of cases where e.g. paths aren't considered as in the same directory when paths are expressed with different separators.
>
> For someone unfamiliar with the tool, is there a simple "smoke test procedure" I could try out with it to kick the tyres?

There's `clangd --check=some_file.cc` which simulates opening up a file in an editor and clicking around in it.
It does require a `compile_command.json` in a parent directory to know about build flags.
To try it out for real you need to connect an editor to it with a plugin (like vscode-clangd).

But realistic problems with paths tend to come from interactions with external sources.
For example, we had a bug that involved:

- vscode sending paths like `file://c:/test.cc` (lowercase C), which made its way into our AST cache
- cmake creating `compile_commands.json` like `C:\test.cc` (uppercase C), which made its way into our index
- when renaming a symbol, results from the two weren't deduplicated against each other so the edit was applied twice

> Anyway, I'll have a look at the seemingly odd/fragile change and get back to you on that.

Thanks!


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D113268/new/

https://reviews.llvm.org/D113268