add normalize and prepare #32
Conversation
This brings in parity with `read-package-json-fast`, which is a normalization method intended for parsing package.json files inside of an existing node_modules. Tests were copied straight from that package to ensure functionality was compatible. The only tests that weren't copied were the ones testing down into the bin normalization. That is delegated to a subdependency so these tests only ensure that *some* normalization is happening. Eventually as we consolidate our package.json reading libs, bin normalization can live in this package and be tested here. Finally, the errors that this package was throwing now include the metadata from the original errors (such as code) instead of making new errors with no context.
wraithgar
force-pushed
the
gar/prepare
branch
3 times, most recently
from
e2d289d
to
3cb844b
Compare
This implements most of the logic found in `read-package-json` with the exception of parsing package comments from specific files. It also shares all the normalization logic into one method, that `normalize` and `prepare` then pick and choose which ones to use. This will allow us to maintain the functionality that `read-package-json` has of skipping certain steps, but doing so in a more config-based way instead of passing in functions to run. All the existing `read-package-json` tests were ported over, and new ones added to get to 100% coverage. For now the 'steps' will be undocumented.
|
File permission on delete in Windows tests, reran the test and it is fine now. Probably a race condition in the test. |
|
flaky windows tests is fixed by |
There was a problem hiding this comment.
This is exciting! I wonder if the bin normalizers should be rolled into this.
This is good, and ready to ship, but I do have some questions.
Would steps be more performant as a Set?
Or should steps map to a scoped object of functions, as an array of steps implies order, and iterating through the steps would make sense.
There's a little bit of a smell repeatedly doing searches on array.
|
bin normalizers could only be rolled in if we exported them separately, lots of things use that one separately.
|
|
One of the lessons learned in the recent ssri/semver optimizations is assumptions are usually wrong, and the only optimizations that should be added are those that got a benchmark and showed a measurable improvement. |
Adds a
normalizeandpreparefunction, to replaceread-package-jsonandread-package-json-fastAll the steps can be omitted by an optional parameter. All of the functionality from both libs was ported except for parsing package comments from specific files (the index.js default is still there).
The
stepsare intentionally undocumented because we may have to tweak them as we find the line between whatnpm pkg fixshould do and what the defaults are.