`clang-format` behaves differently on pipe than with explicit path

[emilyyyylime]

Previous discussions: clangd/clangd#1926, helix-editor/helix#9563

To reproduce:

$ touch .clang-format
$ printf '#include "a.hpp"\n#include "b.hpp"\n' > b.cpp
$ clang-format --style=file b.cpp
#include "b.hpp"
#include "a.hpp"
$ cat b.cpp | clang-format --style=file
#include "a.hpp"
#include "b.hpp"

[JonChesterfield]

From the linked discussion I think the idea is to put b.hpp at the top of the list when formatting a file called b.cpp.

When piped to clang-format, the input file is stdin. I don't think there's any way for a program reading from stdin to guess a corresponding filename for those bytes. Is the intended fix to add an argument to clang-format to tell it what the filename would have been, were it not from stdin?

[emilyyyylime]

This makes sense. It seems I haven't thought this through all the way. Closing in favour of a downstream issue in Helix.

[nv-quan]

For those who end up here from Google, you can pass the filename through --assume-filename option so that they can behave consistently. This is useful when you do something like filtering the file content inside an editor.