No way to use filetype rules when piping with stdin

[orhun]

Repro case:

# typos.toml
[type.md]
extend-ignore-re = ["transfered"]

# file.md
transfered

$ typos file.md # works

$ typos --config typos.toml file.md # works

$ echo "transfered" | typos - # does not work

`transfered` should be `transferred`

$  echo "transfered" | typos --config typos.toml - # does not work

`transfered` should be `transferred`

I looked around in the codebase but couldn't figure out what causes stdin to be handled differently.

[dnaka91]

Just stumbled upon the same issue when adjusting my git-cliff config to integrate typos as a post-processing step (thanks @orhun for git-cliff).

What I found out with the --file-types flag is, that the type of a file is probably derived from the file extension. When using stdin it is always reported as -.

Meaning typos doesn't detect the file type from its content and probably needs an extra flag to define the file type when passing in the content through stdin.

[orhun]

Thanks for looking into this @dnaka91!

Indeed, file type isn't set when the content is given via stdin. I applied the following hacky patch to set it to md to see if it is going to make any difference:

diff --git a/crates/typos-cli/src/policy.rs b/crates/typos-cli/src/policy.rs
index e8fb2e9..e7b7d32 100644
--- a/crates/typos-cli/src/policy.rs
+++ b/crates/typos-cli/src/policy.rs
@@ -330,7 +330,7 @@ impl DirConfig {
                 );
                 self.default
             });
-        (name, config)
+        (Some("md"), config)
     }
 }

Unfortunately it didn't change anything. Maybe it should be set in a different way or we're not exactly on the right track.

[epage]

@orhun that only affected the returned name, it is still using the default config.

btw if you enable logging, the above log line shows up saying that it chose the default policy because an unknown file type was selected.

Could you describe the use case for what you are doing? For example, an extra flag for setting the file type means you need to do the file type detection. Would you instead want to tell us what the effective file name is so we still do our own detection? If that is the case, why can't we be given a file (I'm assuming its in memory?)?

[orhun]

The use case is fixing the the typos for the generated changelog via git-cliff as shown below:

[changelog]
postprocessors = [
  # Check typos with https://github.com/crate-ci/typos
  # If the spelling is incorrect, it will be automatically fixed.
  { pattern = '.*', replace_command = 'typos --write-changes -' },
]

This is handled by piping the output to the given command's stdin, the implementation is here: https://github.com/orhun/git-cliff/blob/7ae77ff0e0a22b5f5e42737204cbf0ab8680f9d7/git-cliff-core/src/command.rs#L48

As you might guess the problem is that the --config does not work like this:

{ pattern = '.*', replace_command = 'typos --config typos.toml --write-changes -' }

[orhun]

Is there any other info that I can provide to debug this further? 👀

[epage]

The question isn't too clear to me. What is being debugged by whom?

[orhun]

Sorry for not being clear. I'm trying to figure out what causes stdin to be handled differently in this case (for potentially fixing the issue) so I need some guidance. Let me know if I can provide any other information about my use-case for figuring out the next steps for making it possible.

[epage]

The difference is that there is no filename to do file type detection from.

Our options for solutions are

• File type detection which would be iffy at best
• A flag to receive a specific file type from --type-list
• A flag to receive the file name. This could also be used in any messages.

Based on your workflow, it seems like either style of flag would work (--type md vs --stdin-file-name CHANGELOG.md).

I would be curious if there is any prior art around this; if there are other tools that need to know the file type for stdin how they handle it.

[orhun]

I'm not aware of any other tools. I think --type md flag would work perfectly in this case.

[epage]

Someone just pointed out that bat has --file-name <name> Specify the name to display for a file.. That would also avoid callers having to figure out the file type (in your case you know it, but it could be more difficult in others).

I am open to merging a PR for a --file-name parameter that (1) is used for file type detection and (2) is shown on the screen.

[orhun]

I'm interested in making a PR but I'm not sure where to start. I added the command-line argument - as the next step should I override the file name if the argument is present? Can you provide some guidance about how to do that?

[epage]

For changing the selected policy, walk_entry needs to know this for determining the file name to use for engine.policy(lookup_path). Not seeing an obvious general solution to this so we'll probably just need to add another parameter to the call chain to include an override_name: Option<&Path>. We should only set this if the flag is used and only allow the flag to be used when reading from stdin.

For showing the name in the messages, we could

• Rename report::MessageStatus to be more general
• Tell it to use the file name for - (except maybe for json output?)
• Overwrite the reported file name with this one.

[epage]

Saw that rustfmt is having a discussion on this to. It seems the discussion is in rust-lang/rustfmt#5266