mirror of
https://github.com/sourcegraph/sourcegraph.git
synced 2026-02-06 20:11:54 +00:00
79226578ec
Right now, we use the default scanner split function, which is `bufio.SplitLines`. This removes the trailing newline and any preceding `\r` before it, then returns the remainder of the token. `PipeOutput` then puts the newline back by invoking `fmt.Fprintln`. This has the positive effect that we essentially normalise `\r\n` and `\n` into `\n`, but the negative effect that every line of output becomes newline-terminated, even if it's a final line that doesn't include a newline. This matters because this output is then used for output variables when executing batch specs, which means that something like `echo -n hi` will end up with an extraneous newline, even though `echo -n` has only written two bytes to stdout. Instead, let's set the split function to one that splits by line, but _retains_ the newline, and then we can write it to the piped writer(s) with `fmt.Fprint` directly. The one tradeoff here is that I haven't reimplemented the `\r` swallowing that `bufio.SplitLines` does. In practice, I don't believe we ever actually used this behaviour: Docker _may_ return extraneous `\r` bytes when running with TTY mode enabled, but we never enable that when executing batch specs, and if the user's container actually outputs `\r` we should actually retain it, rather than trying to filter their output. Fixes #36562. |
||
|---|---|---|
| .. | ||
| logger.go | ||
| pipe_test.go | ||
| pipe.go | ||