search

ambuda-org / vidyut

Infrastructure for Sanskrit software. For Python bindings, see `vidyut-py`.
48 stars 21 forks source link

WASM build #19

Closed shreevatsa closed 1 year ago

shreevatsa commented 1 year ago

Just a demo of derive_tinantas for now: can do similar with derive_subantas and derive_krdantas later.

image

The "Vidyullekhā" name for the demo page is just a small joke, denoting a flash of lightning, and hinting at being ullekha of Vidyut :P

shreevatsa commented 1 year ago

Note: There was some concern about whether the changes for the wasm bindings would make things heavy, so I've put everything under a feature flag in the latest commit 3867b0e4c927209eb23779abb18d1ac18bd7a0cc. It does make things a bit uglier so it's worth thinking about whether it's worth it. For that reason, a comparison of compilation time:

✗ hyperfine --prepare "cargo clean" --cleanup "cargo clean" "cargo build --release" "cargo build --release --features=wasm_bindings"
Benchmark 1: cargo build --release
  Time (mean ± σ):     13.126 s ±  0.188 s    [User: 77.690 s, System: 4.831 s]
  Range (min … max):   12.941 s … 13.478 s    10 runs

Benchmark 2: cargo build --release --features=wasm_bindings
  Time (mean ± σ):     13.561 s ±  0.217 s    [User: 85.318 s, System: 5.679 s]
  Range (min … max):   13.351 s … 14.088 s    10 runs

Summary
  'cargo build --release' ran
    1.03 ± 0.02 times faster than 'cargo build --release --features=wasm_bindings'

And comparison of file sizes:

# ls -lS ../target/release/
# cargo build --release --features wasm_bindings >/dev/null 2>&1 && ls -lS ../target/release/
11485584 11956672 libvidyut_prakriya.rlib
 2151211  2158315 test_tinantas
 2150973  2157725 create_tinantas
 2140445  2147517 explain_tinanta
 2081789  2085821 create_krdantas
 2060363  2065003 test_krdantas
 2043165  2046973 explain_krdanta
 1796781  1798605 explain_subanta
 1437950  1445022 create_test_file
 1172907  1174843 test_subantas
   16741   433749 libvidyut_prakriya.dylib
    3120     3182 create_test_file.d
    3118     3180 create_krdantas.d
    3118     3180 create_tinantas.d
    3118     3180 explain_krdanta.d
    3118     3180 explain_subanta.d
    3118     3180 explain_tinanta.d
    3114     3176 test_krdantas.d
    3114     3176 test_subantas.d
    3114     3176 test_tinantas.d
    3050     3112 libvidyut_prakriya.d
shreevatsa commented 1 year ago

Demo currently (temporarily) serving at https://profound-croquembouche-79b4a5.netlify.app/

akprasad commented 1 year ago

For now I've decided to just merge this into the repo. PR looks good to me -- just to check, are there any other changes you want to make, or is it ready to merge?

akprasad commented 1 year ago

(Another reason to merge it in as-is: using the command line to look at prakriyas is really tedious, so I'll probably use this as a debugger!)

shreevatsa commented 1 year ago

Don't have any other changes as of now, so feel free to just merge.

On Sat, Dec 31, 2022, 19:40 Arun Prasad @.***> wrote:

(Another reason to merge it in as-is: using the command line to look at prakriyas is really tedious, so I'll probably use this as a debugger!)

— Reply to this email directly, view it on GitHub https://github.com/ambuda-org/vidyut/pull/19#issuecomment-1368341228, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAF3MKWQWKFUPS7VGOEGPKDWQD4CVANCNFSM6AAAAAATNGKCBQ . You are receiving this because you authored the thread.Message ID: @.***>

akprasad commented 1 year ago

GitHub did a merge commit by default, so I edited this commit to be a squash and force-pushed as 4b50c14.

shreevatsa commented 1 year ago

Some (already known) possible follow-up items (let me know if these should be tracked as an issue or even multiple issues):

Frontend at https://ambuda-org.github.io/vidyullekha/

Code:

akprasad commented 1 year ago

Yes, this all looks right to me. In terms of priority, I would rank these as:

  1. separate para/atma -- must be fixed before sharing out
  2. Add search box for dhatus -- for less tedious inputs
  3. nice URLs -- for easier sharing
  4. side by side -- nice to have but a little trickier on the frontend side.
  5. show final form at the end -- tricky to get right if associating it with the last applied rule. Otherwise it's a simple append.

Let's file separate tickets for these. I'll create tickets for 1 and 2, and I'll start work on 1. Can you create tickets for the others?

For wasm -- sounds good. For now, let's use the wasm macro only on simple enums. I wonder if we can get a nicer API by allowing it on everything args -- let me know if you're curious to try it out.

For debug -- right now our goal is overall debugging, so feel free to turn on whatever makes sense.

shreevatsa commented 1 year ago

I see (1) is filed as https://github.com/ambuda-org/vidyut/issues/20 and (2) as https://github.com/ambuda-org/vidyut/issues/21. So for the rest:

I'll look into just sending a PR for removing the separate wasm_bindings feature, and if it doesn't turn out to be trivial, create a ticket. The debugging from Chrome Devtools we can probably look into later.