Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Don't use matcher from a .gitignore against includePatterns
When attempting to determine whether a .gitignore should be respected, we previously tried to see, whether any of the patterns in the .gitignore match against the includePatterns. As @fsoikin noticed [here](https://github.com/purescript/spago/pull/1296/files/b84981dd086219bc1157a440667aca92ea23efb4#r1815996597) this does not really make any sense in case the includePatterns really are actual patterns and not just paths. Therefore, instead of matching the .gitignore patterns against the includePatterns directly, we now only match them against the base of each includePattern. The base of a pattern is always a path. However, some includePatterns may not have a base, for example when searching for `**/foo/bar`. What should one do here? I think disrespecting .gitignores is not really feasible in such a case, because how can we tell if something that the .gitignore excludes could match the baseless pattern `**/foo/bar` without actually walking the excluded directories? So for example, if there was a file: `./abc/foo` and a .gitignore with the pattern `abc` and we glob for `**/foo`, then we do not find `./abc/foo`, because the .gitignore was respected. I think, this is the behaviour we would want anyway, we only ignore a .gitignore, when it is obvious that the glob target is necessarily within an ignored directory. Like globbing for `.spago/p/arrays-7.3.0/src/**/*.purs` when `.spago` is gitignored. This was already the behaviour previously. In the above example, we would have previously matched `abc` against `**/foo`, which would be false and therefore no conflict would have been dectected, so the .gitignore would also have been respected. This PR only makes this behaviour explicit and removes the, incidentally working, match of a pattern against another pattern. The new test also passes without this change, but i think it's a nice kind of documentation. I wonder, whether the globbing code should even attempt to disrespect .gitignore files, doesn't the calling code always know, whether it should limit its search to the non-ignored files? For example, when globbing for the source files of a package in .spago, we know beforehand, that a nonGitignoringGlob (otherwise known as `glob` :^D) would suffice.
- Loading branch information