GSOC progress update #2: Extracted libraries & Hoogle
More progress updates on my GSOC project. Things seem to be going well!
Extracted Haskell libraries
I extracted two Haskell libraries out of what I’ve been doing recently.
The first, aeson-better-errors, arose from a desire to get better errors out of aeson parsers for common errors, such as type mismatches, missing object keys, and so on. In particular, I wanted to know exactly where in the JSON the problem is, since this makes diagnosing some problems significantly easier; for example, bugs in serializers, or attempting to parse data using an older serialization format.
Hopefully an example will make it clear what I mean:
deeplyNested :: Parse e Int
deeplyNested = key "a" (nth 3 (key "b" (key "c" asIntegral)))
λ: printErr $ parse deeplyNested "{\"a\":[null,null,null,{\"b\":{\"c\":null}}]}"
At the path: ["a"][3]["b"]["c"]
Type mismatch:
Expected a value of type number
Got: null
One of my aims for the tooling around Pursuit, and particularly the process of submitting your packages, is that it should be as pleasant as possible. We want people to use it, after all! I created aeson-better-errors as a part of this aim.
You can find aeson-better-errors on Hackage. I also wrote a tutorial/introduction blog post.
The other library, bower-json, is a little less interesting. It simply provides
a data type and ToJSON/FromJSON instances for Bower’s package manifest file
format (that is, bower.json files). You can find bower-json on Hackage
too.
Changes to the psc-publish tool
psc-publish is the tool which looks through all the source code of a
PureScript package, extracts the information required to host it on Pursuit,
and dumps that information as JSON. Previously, this included rendering code;
that is, taking parts of a PureScript AST and converting them into an
intermediate format called RenderedCode, suitable for generating plain text
or highlighted HTML from. Now, the rendering has been delayed until later, and
so it is no longer a responsibility of psc-publish.
This means that the information stored in the JSON output by psc-publish now
includes values of types such as Language.PureScript.Type and
Language.PureScript.Kind in place of where it previously would have had
RenderedCode, and this gives us some more flexibility with respect to what we
do with it.
The purpose of this change is to allow transformations on types and kinds inside the Pursuit server, in order to output Hoogle files which can be understood as Haskell; for the time being, this is the approach we’re pursuing, since the type systems of Haskell and PureScript are very similar. For example, a type synonym such as the following:
type Person a = { name :: String, age :: Int | a }
would currently be encoded like this (although this is very much a first pass):
type Person a = Object (PS_Row (PS_Label_name String) (PS_Label_age Int) (PS_Row_Tail a))
Once we have things working, I will probably investigate whether other encodings are more effective.
The Pursuit server
Phil Freeman has now deployed a pre-alpha version of pursuit to http://new-pursuit.purescript.org. Feel free to have a look around, but please don’t start trying to actually use it just yet. Additionally, trying to actually deploy it has revealed various deficiencies; generally, deployment seems harder than it ought to be. For example, setting configuration values such as the port to listen on, or the data directory, or OAuth tokens for the Github API is a bit painful. There’s a weird mixture of parsing and using configuration settings at both compile time and run time that I haven’t fully understood, and (as unfortunately seems to often be the case with compile-time code) the errors are not very clear. I’m planning on addressing this soon.