py/persistentcode: Only store a single instance of source_file in mpy. #4472
Conversation
On average reduces .mpy file size by about 5%.
|
For background see #3054 |
Well, one obvious optimization to would be to allow to merge several .mpy's into one. But to get real benefits from that, it would take to allow to share e.g. the same constant table among the different functions, i.e. change serialization format anyway. But if planning to do that, how it should be done is apparently introducing a bitflag field before each code object with bits of "read new/use the same as previous" for filename/const table/etc. |
I'm not sure how that would work: a file has meaning to Python import ( |
Apparently, that would need to be addressed somehow by someone doing that optimization, e.g. init code added which creates entries in sys.modules manually, etc. Conceptually, it's the same thing which Android's .dex did to individual Java .class'es. |
|
Superseded by #4549 |
This is a relatively simple change to only store a single instance of the source file name in a .mpy file. It assumes that all code blocks (classes, functions, lambdas, etc) in a single .mpy file come from the same .py source file, which I think is a fair assumption (and is the case for the output of mpy-cross).
No extra RAM is needed to save or load the new .mpy files. It should be faster loading because the source file is only loaded at the start of loading the .mpy.
On average it reduces .mpy file size by about 5%.