bpo-40280: Add Tools/wasm with helpers for cross building (GH-29984)

[tiran]

https://bugs.python.org/issue40280

[tiran]

$ CONFIG_SITE=../../Tools/wasm/config.site-wasm32-emscripten \
  emconfigure ../../configure -C \
    --host=wasm32-unknown-emscripten \
    --build=$(../../config.guess) \
    --with-build-python=$(pwd)/../build/python
$ emmake make -j$(nproc) python.html
$ ls -s --si python.*
3.0M python.data
107k python.html
361k python.js
6.6M python.wasm

[emmatyping]

This will definitely make experimenting with Emscripten based CPython a lot easier!

I have a few suggestions/comments but overall this looks good.

[emmatyping]

The binary size looks really great! Though with the number of things disabled, it will probably grow over time...

[tiran]

With --disabled-test-modules, without decimal unittest xml and without some C extension

# Modules/Setup.local
*disabled*
_decimal
pyexpat
_elementtree
_lsprof
audioop

I'm able to shrink the core to about 6 MB

# du --si -s -c python.*
2.7M    python.data
107k    python.html
275k    python.js
3.3M    python.wasm
6.4M    total

The python.wasm file can be compressed well. If we assume that all browsers support transparent compression, then we are down to about 4 MB with room for more removals.

# du --si -s -c python.*
2.7M    python.data
107k    python.html
275k    python.js
1.1M    python.wasm.gz
4.2M    total

The largest 25 files are:

# unzip -l usr/local/lib/python311.zip  | sort -n | tail -n 26
    57953  2021-12-13 19:58   threading.pyc
    59930  2021-12-13 19:59   optparse.pyc
    61650  2021-12-13 19:58   subprocess.pyc
    61916  2021-12-13 19:56   collections/__init__.pyc
    63723  2021-12-13 19:56   enum.pyc
    63980  2021-12-13 19:59   pickle.pyc
    66135  2021-12-13 19:59   pdb.pyc
    67936  2021-12-13 19:58   locale.pyc
    68530  2021-12-13 19:58   asyncio/base_events.pyc
    70300  2021-12-14 09:56   _sysconfigdata__emscripten_.pyc
    71134  2021-12-13 19:59   difflib.pyc
    77197  2021-12-13 19:58   datetime.pyc
    77293  2021-12-13 19:59   ipaddress.pyc
    78354  2021-12-13 19:56   ast.pyc
    79675  2021-12-13 19:59   pickletools.pyc
    82993  2021-12-13 19:58   logging/__init__.pyc
    85730  2021-12-13 19:56   zipfile.pyc
    86382  2021-12-13 19:59   mailbox.pyc
    86537  2021-12-13 19:56   argparse.pyc
    88029  2021-12-13 19:59   tarfile.pyc
    92877  2021-12-13 19:59   doctest.pyc
   108300  2021-12-13 19:56   inspect.pyc
   109640  2021-12-13 19:58   typing.pyc
   118759  2021-12-13 19:59   pydoc.pyc
   253750  2021-12-13 19:59   html/entities.pyc
  5711264                     347 files

@brettcannon now is a good time to discuss what modules belong in a minimum stdlib.

[emmatyping]

Wow that's a great reduction in size! I played with thin lto and for some reason it greatly increases the binary size (to 21MB!). I guess inlining makes the build a lot bigger. One other idea I had which gives small gains in binary size is compiling with -Os, which seems to not make much of a difference vs -Oz.

[tiran]

Actually it is less. I think I introduced a regression in GH-30001. makesetup is dropping too many modules from the final binary.

[brettcannon]

Suggested change

	tool chain and a build Python interpreter.
	tool chain and a built Python interpreter.

[emmatyping]

build here is I believe a reference to the autotools/configure --build argument. (Same for line 11 below).

[tiran]

That is correct. In lack of a better term I introduced the term "build Python"/

[brettcannon]

My brain just keeps reading "build" as a verb instead of an adjective. 😄