Make Sourcemap `upload` and `inject` `--ignore` paths more intuitive

[szokeasaurusrex]

Currently, paths passed to the sourcemaps upload and sourcemaps inject subcommands' --ignore argument are always interpreted as being relative to the path(s) being uploaded/injected. In the case where multiple paths are passed to these commands, --ignore ignores the specified ignore path in both of the upload/inject paths.

Current behavior example

Suppose I run the following command:

sentry-cli sourcemaps {inject|upload} path_a path_b \
    --ignore path_a/test \
    --ignore /Users/myuser/Development/my_project/path_a/exclude_files

We will inject/upload sourcemaps in ./path_a and ./path_b, ignoring any files in the following paths:

• ./path_a/path_a/test
• ./path_a/Users/myuser/Development/my_project/path_a/exclude_files
• ./path_b/path_a/test
• ./path_b/Users/myuser/Development/my_project/path_a/exclude_files

In this example, the behavior is likely not what the user expected.

New behavior

The command above should inject/upload sourcemaps in ./path_a and ./path_b, ignoring any of the files in the following paths:

• ./path_a/test
• /Users/myuser/Development/my_project/path_a/exclude_files

In other words, we should stop interpreting the --ignore paths as being relative to the paths to inject/upload. If the paths start with a /, we interpret them as absolute paths; otherwise, we should interpret them as being relative to the current directory (i.e. ./).

[linear]

CLI-228

[szokeasaurusrex]

As we support globbing, and the current globs have .gitignore semantics, fixing this behavior seems less trivial than I had hoped. Any fix would likely alter semantics in complex ways that could break edge cases that folks might be relying on.

So, I think the better path here will be to, first of all, document the current behavior clearly, and second, we could consider introducing a new ignoring mechanism which supports absolute paths; as that would be separate, we would not need to worry about breakages.