MAIN POINT: It's too big and too slow, but I want to get back to oil. It can be made smaller and faster AFTER OIL.
SECOND POINT: Blog themes for LATER. After Boil.
What I described in the last post is now working:
I still haven't hooked up the opy front end, but I've done enough tests.
The mai issue is that it's too slow. I noticed that the compiler2 generates bytecode that is worse than CPython. It was removed from Python 3 because it was not maintained that well.
However, it does work, and gives me a lever to FORK the language, so I still plan to keep the opy front end.
Not sure how detailed I want to be on this.
Maybe just mentrion metrics.
- fix makefile bugs
- improve on all these metrics, in order of priority:
- startup time
- binary size -- vm size, etc. with bloaty
- lines of code in the tarball
- build time -- right now it's a single C compiler invocation. It's fast
on my machine. Might want to move to separate compilation.
- it's OK on my machine, but it will be slow on other machines because:
1. Clang is faster than gcc
1. My machine is fast
- benchmark speed -- e.g. parsing with wild.sh.
- it will likely slow down due to the code generated by OPy.
- add coverage, ASAN back
- test the tarball on different systems.
- deal with ./configure -- freeze it in a container
- integrate Opy compiler. Right now it's using CPython compiler. This
could fix the stack trace issue.
- speed benchmarks
- add back PGO -- Python has it.
- test it against Python.
see Lessons Learned in oil-sketch repo
configure script. Generate header file and make flags.