search

atom / fuzzy-finder

Find and open files quickly
MIT License
274 stars 138 forks source link

Consume STDERR from ripgrep to allow indexing process to complete #406

Closed fredden closed 4 years ago

fredden commented 4 years ago

Description of the Change

When indexing a project with a lot of broken symbolic links, the indexing process never completes. This is because the ripgrep process has written more data to its error output (STDERR) than the calling process has buffer to receive; until this buffer is cleared, the process will not terminate. The solution to this is to process the data on the receiving side.

This pull request adds a simple no-op handler for anything sent back on STDERR from ripgrep.

Alternate Designs

I considered logging the error output from ripgrep. This produced a lot of noise when running the test-suite and is unlikely to be of use for anyone in the real world. Logging the error messages returned from ripgrep is a trivial change.

Benefits

Indexing process always completes.

Possible Drawbacks

Diagnosing some issues may be difficult due to the lack of logging, however before this change any messages sent to STDERR were already being lost.

Applicable Issues

No obviously related open issues in this repository at time of opening this pull request. Lots of seemly-related external results when searching for this problem on the web.

rsese commented 4 years ago

Thanks for the contribution! CI failure looks unrelated so we'll ask the other maintainers to take a look.

darangi commented 4 years ago

Thanks for the contributions 🙇 @fredden

BurntSushi commented 4 years ago

Small note: if you are dropping stderr anyway, then you can use --no-messages (and possibly --no-ignore-messages) to tell ripgrep to be more quiet about file system related errors (and errors from parsing gitignore files).