pj/sanderling
1
0 Fork 0
mirror of https://github.com/priyanshujain/sanderling.git synced 2026-10-02 19:17:10 +00:00
Code Issues Packages Projects Releases Wiki Activity

docs(ci): explain why a submit tap landing on home is the bug

Browse Source
This commit is contained in:
pj committed 2026-08-13 22:41:24 +05:30
1 parent 7cd3bc7749
commit 2f782608c5
1 file changed
+15 -9
Download Patch File Download Diff File Expand all files Collapse all files
docs/development/ci.md
+15 -9
View File
@@ -27,7 +27,7 @@ script is plain bash so you can reproduce a job locally:
SEED=3 MAX_STEPS=240 .github/scripts/folio-run.sh android SEED=3 MAX_STEPS=240 .github/scripts/folio-run.sh android
``` ```
**android and ios expect the bug.** Folio double-submits a transaction when the **Every leg expects the bug.** Folio double-submits a transaction when the
submit button is double-tapped, and `submitMovesBalanceByTypedAmount` is the submit button is double-tapped, and `submitMovesBalanceByTypedAmount` is the
property that catches it. The runs pass `--exit-on-violation`, so: property that catches it. The runs pass `--exit-on-violation`, so:
@@ -40,12 +40,16 @@ property that catches it. The runs pass `--exit-on-violation`, so:
Distinguishing 0 from 1 is the whole point of the exit code: a job that only Distinguishing 0 from 1 is the whole point of the exit code: a job that only
knew "non-zero" could not tell a working fuzzer from a broken emulator. knew "non-zero" could not tell a working fuzzer from a broken emulator.
**web is a health gate**, not an expect-the-bug leg. The same spec logs into the On web the property fires for a structural reason worth knowing when you read a
wasmJs build and drives it to the transaction screen (the job asserts the trace witness. `AddTransactionViewModel.submit()` inserts the transaction and pops one
reached `AddTransactionScreen`), but the submit property cannot fire there: it level, so a submit tap lands on Ledger; reaching Home takes two pops, which takes
keys off `state.lastAction`, which the web runtime reports as `null`, and off the two inserts. A submit tap that lands on Home therefore is the double submit. Over
action's selector, which the web picker does not carry, since it emits 451 single taps on the submit button, none reached Home.
coordinates and element identity never crosses the V8 boundary.
The witness's arithmetic is noisier than that reasoning. `totalBalance` only
refreshes on a Home step, so the recorded delta covers every transaction since
the last Home visit, and only some witnesses read as a clean 2x the typed amount.
The violation is still real; the number beside it just needs that context.
The wasmJs app is served with `Cross-Origin-Opener-Policy` and The wasmJs app is served with `Cross-Origin-Opener-Policy` and
`Cross-Origin-Embedder-Policy` headers, because its sqlite worker needs `Cross-Origin-Embedder-Policy` headers, because its sqlite worker needs
@@ -54,8 +58,10 @@ every step observes an empty accessibility tree.
The seeds in the workflow are calibrated, not guessed. On an M-series mac, The seeds in the workflow are calibrated, not guessed. On an M-series mac,
android seed 3 finds the bug at step 110-116 (seeds 1, 2 and 4 run 120 steps android seed 3 finds the bug at step 110-116 (seeds 1, 2 and 4 run 120 steps
clean) and ios seed 1 finds it at step 129-134, both against a 240-step budget. clean), ios seed 1 finds it at step 129-134, and web seed 1 finds it at step 109,
The web leg ran five seeds x 200 steps clean. reproducing 3 runs out of 3. All three run against a 240-step budget. Keep the
web seed pinned: 9 of 12 random seeds found the bug within 200 steps, so an
unpinned one would flake.
The ios leg passes `--clear-data=false`, because the job installs a fresh build The ios leg passes `--clear-data=false`, because the job installs a fresh build
immediately before the run and a freshly installed app is already clear state. immediately before the run and a freshly installed app is already clear state.