Repository navigation
Conversation
…fies
Every `difficult` run currently prints
Risch improvements vs baseline (engine got better — tighten
test/difficult_baseline.jl on this PR):
[3] sin(x) expected=1 actual=0
[26] sin(x)*(x^2) expected=1 actual=0
[27] cos(x)*(x^3) expected=1 actual=0
[49] sin(5x)*(2 + 3x) expected=1 actual=0
[55] sin(1 / x) / (x^2) expected=1 actual=0
Those five now verify symbolically against the reference antiderivative instead
of returning an unverifiable equivalent form. They show up identically on all
nine `difficult` jobs (Julia 1.10, 1 and pre across ubuntu, macos and windows)
of several unrelated pull requests, so this is dependency drift rather than any
one change, which is why it is refreshed on its own here rather than folded into
a feature branch.
Only the Risch expectation moves, from 1 to 0; the RuleBased column is
untouched. The file stays at 177 entries, and it is indexed by position, so the
spot checks [15] `(0, 2)` and [158] `(1, 2)` are unchanged.
Assisted-by: AI
|
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #148 +/- ##
==========================================
+ Coverage 50.98% 51.19% +0.21%
==========================================
Files 23 23
Lines 4309 4213 -96
==========================================
- Hits 2197 2157 -40
+ Misses 2112 2056 -56 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
s-celles-bot
added a commit
to s-celles/SymbolicIntegration.jl
that referenced
this pull request
Oct 9, 2026
CI reported one RuleBased regression on this head, `[158] sqrt(2 - x - x^2)/x^2`, expected 0 and got 1, on ubuntu, macos and windows alike. It is not a regression of the engine. The residual `r - ref` carries floats — `0.35355339059327373` and `2.8284271247461903`, that is `1/(2*sqrt(2))` and `2*sqrt(2)` — and they come from the reference value, not from the result: in this corpus `sqrt(2)` is Julia's `sqrt` on an `Int`. The `SymbolicUtils = "4.49"` bound that JuliaSymbolics#136 brings into this branch makes the integrator's own output exact, so an exact result now faces an inexact reference and `simplify` cannot cancel the difference. The entry verified on the earlier head only because both sides were inexact. This is the same defect this PR already fixes for four other Apostol entries, and 158 is a fifth that was missed. Its two bare `sqrt(2)` become `sqrt(Symbolics.Num(2))`, the public spelling used for the others — `Symbolics` is reachable from `lib/SymbolicIntegrationMaxima`, which reads the same corpus, whereas an internal helper would not be. Verified: the reference is now free of floats, `simplify(r - ref; expand = true)` is `0`, and the harness scores the entry 0 again, so the baseline stays `(0, 2)`. `TEST_GROUP=difficult`: exit 0, **no regressions**, RuleBased up from 93 to 94 successes, Risch unchanged at 58 with 0 errors. The improvements the run still lists are left alone: entry 132 keeps its tolerant value, which its own comment explains returns code 2 on Julia pre and Windows, and the five Risch entries belong to JuliaSymbolics#148. Assisted-by: AI
Brings in JuliaSymbolics#147, which fixes baseline entry 15, `sin(sqrt(1+x))/sqrt(1+x)`. That was the single RuleBased regression failing the `difficult` gate on this pull request; it came from `main`, not from refreshing the baseline. No conflict. `TEST_GROUP=difficult` on the merged branch: exit 0, **no regressions**, and the five Risch entries this pull request tightens no longer appear as improvements — they now match the baseline, which is the whole point. RuleBased 92 succeeded, Risch 58, 0 errored. The one improvement the run still reports, entry 132, is left alone on purpose: its own comment records that RuleBased returns code 1 on most platforms but code 2 on Julia pre and Windows, so the tolerant value is deliberate and tightening it would break those two. Assisted-by: AI
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Every
difficultrun currently ends with the harness asking for this:Those five integrands now verify symbolically against the reference antiderivative, instead
of returning an equivalent form the harness cannot check (code 1).
They appear identically on all nine
difficultjobs — Julia 1.10, 1 and pre across ubuntu,macos and windows — of several unrelated pull requests (#138, #140, #141, #146, #147), so this
is dependency drift rather than the effect of any one change. That is why it is refreshed on
its own branch instead of being folded into a feature PR, where it would look like part of
that change.
Only the Risch column moves, from 1 to 0. The RuleBased column is untouched. The file is
indexed by position, so it still holds 177 entries and the spot checks
[15] = (0, 2)and[158] = (1, 2)are unchanged.This does not by itself turn
difficultgreen: improvements do not fail the gate, onlyregressions do. The one regression it reports, entry 15, is fixed by #147.
Assisted by: AI
CI on head
4ca132729 pass, 0 fail — and the
difficultgroup is green.Before current main was merged in, this pull request failed nine
difficultjobs. Allnine were the same single RuleBased regression, baseline entry 15,
sin(sqrt(1+x))/sqrt(1+x), which fails identically onmainand which #147 fixes. Mergingmain cleared them without touching a line of this change.