Prior to this change, GetLanguages collected all candidate languages from each
strategy to pass to the next strategy (without de-duplicating them). Linguist
only uses the previous strategy's candidates for the next strategy. Also, it
would overwrite languages with nil if a strategy returned that, so you could get
into a situation where you go from multiple languages to no language.
See the Ruby code for details: aad49acc06/lib/linguist.rb (L14-L49)
This addresses https://github.com/src-d/enry/issues/207 because GetLanguages
should not return all candidates detected, otherwise it would work differently
than Linguist.
The Linguist-defined language IDs are important to our use case because they are
used as database identifiers. This adds a new generator to extract the language
IDs into a map and uses that to implement GetLanguageID.
Because one language has the ID 0, there is no way to tell if a language name is
found or not. If desired, we could add this by returning (string, bool) from
GetLanguageID. But none of the other functions that take language names do this,
so I didn't want to introduce it here.
Closes#17
Implements the IsGenerated helper function to filter out generated
files using the rules and matchers in:
- https://github.com/github/linguist/blob/master/lib/linguist/generated.rb
Since the vast majority of matchers have very different logic, it cannot
be autogenerated directly from linguist like other logics in enry, so it's
translated by hand.
There are three different types of matchers in this implementation:
- By extension, which mark as generated based only in the extension. These
are the fastest matchers, so they're done first.
- By file name, which matches patterns against the filename. These
are performed in second place. Unlike linguist, we try to use string
functions instead of regexps as much as possible.
- Finally, the rest of the matchers, which go into the content and try
to identify if they're generated or not based on the content. Unlike
linguist, we try to only read the content we need and not split it
all unless it's necessary and use byte functions instead of regexps
as much as possible.
Signed-off-by: Miguel Molina <miguel@erizocosmi.co>
