Hit the GSoC "suggested pencils down" date yesterday. That means there's one week left on the project, so I figured it was time for a wider overview.
What's done so far:
- DecNum is done, it provides big decimal floating point numbers. It has inherited some idiosyncrasies from the decNumber library, but that was to be expected.
- The DecNum tests are pretty much done. Most were lifted from the decTest suite and cover the various operations possible on DecNums.
- DecInt is done, it provides a integer-only version of DecNum, and is feature-complete, if a bit under-tested.
- DecInt is under-tested right now as the decTest suite does not provide many 'integer tests' for us to reuse, The basic functionality is partly tested by DecNum, since DecInt is a subclass of that, but I'll try to add some more standalone DecInt tests in the remaining time.
- The decTest parser is done. Ditched the perl script and made a PCT-based parser wich can run the decTests on it's own and output TAP, or emit the equivalent PIR (which we keep checked into the repo to cut down on build time and reduce dependencies).
- All of the 'magic numbers' needed to interface with the low-level library are now defined as ".macro-const" in a .pasm file we install alongside the library. The aux/ directory contains the c file that can be used to re-generate the constants should the library ever change their definitions.
What I plan to do with the week I have left:
- Go over the code once more, with fresh eyes, and clean it up as needed. I don't foresee any functional changes from now on.
- Write more tests for DecInt. It works well, but it still needs more testing.
- Start the much-postponed user-level documentation. I don't expect to have this beyond a draft status by next week, but I'd like to get started anyways.
What didn't get done:
- My original plan included a third PMC, DecRat, that would provide support for rational numbers. Unfortunately the wasn't enough time left to do that, But on the other hand I don't really think the decNumber library is really suited to this usage, If I had to start over I wouldn't try anything more complicated than a "DecFix" PMC (for fixed-precision decimal numbers) without investing time to modify the library.
What I'd like to do post-SoC:
- Write a detailed report with the lessons I've learned and the opinions I've formed during this project. It will probably include some thoughts I have for BigNums on the parrot core.