Open gilch opened 1 year ago
If I do put it in __annotations__
, maybe I should use typing.Annotated. Not available until 3.9 though, so I'd have to upgrade Python again. Might be moot unless I'm already using annotations for types at the module level, which I'm not.
sphinx.ext.viewcode makes it fairly easy to add links to a source page in the API docs, at least if the source is Python. Linking to the compiled Hissp seems less useful, although I haven't even gotten that much working.
I'd need to add a Sphinx extension to do this. (The
conf.py
can also be a Sphinx extension given asetup()
function.) I've already got alissp_lexer
andlissp_directive
. I'd need something likeapp.events.connect("viewcode-find-source", ...
to handle the appropriate event. Experiments prove this can work with non-Python files, although I'm not sure if I can add syntax highlighting.Linking to the appropriate position in the
macros.lissp
source file requires writing a program for tracking the line numbers of definitions. Just compiling the file executes it, so I might as well use the compile time namespace to determine when a top-level definition has happened, rather than try to figure it out statically.The machinery already partially exists to do this. The
Lexer
andLissp
classes have some position tracking for error messages. I've resisted putting that in the compiler, which might be compiling Hissp that hasn't even been read from text. It's not supposed to have to know anything about the reader. I've certainly thought about source mapping for debuggers though. Tuples can't carry metadata, and I don't want to use anything more complicated instead. Not even a tuple subclass. That would violate core design goals of Hissp.However, when examining this problem I realized that metadata could be transmitted from the reader to the compiler out-of-band, at least at the granularity of top-level forms, either through a parameter if the reader is driving, or through a callback. This would take more reworking than I'd like to tackle before the next release, but seems doable.
The metadata could then be emitted as a comment (similar to macroexpansions) or added to
__annotations__
or something like that.