Forward dotgithub-ref to tests.yml; default to master by ChrisRackauckas-Claude · Pull Request #137 · SciML/.github

ChrisRackauckas-Claude · 2026-09-09T10:36:49Z

Summary

After Fix LTS develop_sources for URL+rev [sources] #136, LTS still called the old develop_sources from the floating v1 tag because:
1. tests.yml defaulted dotgithub-ref to v1
2. grouped-tests.yml / sublibrary-project-tests.yml never forwarded dotgithub-ref into tests.yml
Default dotgithub-ref to master and forward it so URL+rev [sources] pins work on Julia 1.10 (SI, Surrogates, Sundials downstreams, etc.).
Please also bump the floating v1 tag (and/or cut v1.29.0) so packages still on @v1 pick this up without pinning @master.

Test plan

Local include("test/runtests.jl") (new forwarding/default assertions pass)
CI on this PR

Made with Cursor

grouped-tests / sublibrary-project-tests never passed dotgithub-ref into tests.yml, so LTS always checked out the stale floating v1 develop_sources (Pkg.develop rejects rev). Default master so SciML#136's Pkg.add path is used until v1 is retagged / a new v1.x release is cut. Co-authored-by: Cursor <cursoragent@cursor.com>

Floating v1 still has develop_sources that Pkg.develop-s rev pins. Co-authored-by: Cursor <cursoragent@cursor.com>

Co-authored-by: Cursor <cursoragent@cursor.com>

ChrisRackauckas-Claude · 2026-09-09T10:38:33Z

Follow-up to #136: packages on grouped-tests.yml@v1 still cannot pass dotgithub-ref through to tests.yml, so LTS kept using the stale floating v1 develop_sources.

After this merges, please retag floating v1 → master and/or cut v1.29.0 so the org does not need temporary @refs/pull/137/head / @master pins (SI#555, Surrogates#611, SSDE#162, SBMLToolkit#245 already pin this PR head for LTS URL [sources]).

Cross-repo reusable workflows cannot use refs/pull/*/head. LTS develop_sources still hits stale v1 until SciML/.github#137 merges and v1 is retagged; Integer/Dict fixes remain. Co-authored-by: Cursor <cursoragent@cursor.com>

Await Sundials#567 release / SciML/.github#137+v1 retag for LTS URL sources. Co-authored-by: Cursor <cursoragent@cursor.com>

ChrisRackauckas-Claude · 2026-09-09T10:45:03Z

Note: packages cannot `uses: ...@refs/pull/137/head` cross-repo (workflow file validation fails). After merge, please retag floating `v1` → this commit and/or cut v1.29.0 so `grouped-tests.yml@v1` callers pick up the forward + default-`master` without per-package pins.

ChrisRackauckas-Claude · 2026-09-09T11:01:47Z

CI is green on this PR. Ready for review/merge when a maintainer has a moment — this unblocks SciML 32-bit CI parity work.

ChrisRackauckas-Claude · 2026-09-09T11:32:36Z

Merge request (32-bit SciML CI parity)

Ready to merge: forwards dotgithub-ref; unblocks LTS URL [sources] via master scripts. Please merge and retag floating v1 / cut v1.29.0.

CI is fully green on this PR.

URL [sources] clones on 32-bit Julia fail with `failed to stat '/etc/gitconfig'` when the system file is missing, and with EEXISTS when a prior partial clone remains. Touch the config and clear clone caches in the i386 setup step before setup-julia/Pkg. Co-authored-by: Cursor <cursoragent@cursor.com>

ChrisRackauckas-Claude · 2026-09-09T11:57:15Z

Merge request (32-bit SciML CI parity) — elevated

This PR now also includes an i686 LibGit2 workaround in tests.yml:

sudo touch /etc/gitconfig (fixes failed to stat '/etc/gitconfig' on URL [sources] clones)
clear ~/.julia/clones (fixes EEXISTS partial clones)

Without this, any package temporarily URL-sourcing an upstream fix cannot validate on x86 CI (Sundials, DiffEqBase, Nemo, MCI, …). Please merge + retag floating v1 / cut a release ASAP.

ChrisRackauckas-Claude · 2026-09-09T12:04:05Z

Friendly bump: fully green. The new i686 /etc/gitconfig + clones cleanup is blocking validation of every temporary URL [sources] pin on x86 (Sundials/DiffEqBase/Nemo/MCI). Please merge + retag floating v1.

Otherwise the i686 /etc/gitconfig + clones cleanup in tests.yml is never executed when packages consume this PR tip. Co-authored-by: Cursor <cursoragent@cursor.com>

After merge, master carries the i686 gitconfig/clones workaround; packages on grouped-tests@master then execute that tests.yml tip. Co-authored-by: Cursor <cursoragent@cursor.com>

* Temporarily source Sundials 32-bit NVector fix for CI. Co-authored-by: Cursor <cursoragent@cursor.com> * CI: use SciML/.github#137 so LTS can Pkg.add URL [sources]. Co-authored-by: Cursor <cursoragent@cursor.com> * Revert #137 workflow pin; drop temporary Sundials [sources]. Await Sundials#567 release / SciML/.github#137+v1 retag for LTS URL sources. Co-authored-by: Cursor <cursoragent@cursor.com> * Re-source Sundials 32-bit NVector fix; CI on SciML/.github@master. Temporary until Sundials#567 merges and registers. Downgrade disabled while URL [sources] are present (LTS manifest rewrite drops them). Co-authored-by: Cursor <cursoragent@cursor.com> * Retest against Sundials tip without DiffEqBase [sources]. Co-authored-by: Cursor <cursoragent@cursor.com> * CI: retry x86 after i686 git clone flake on Sundials [sources]. Co-authored-by: Cursor <cursoragent@cursor.com> * Also source DiffEqBase 32-bit interp_points fix. Pin DiffEqBase directly (ODE#4487) alongside Sundials so nested [sources] subdir resolution is not required on julia-pre / i686. Co-authored-by: Cursor <cursoragent@cursor.com> * Drop DiffEqBase [sources]; rely on Sundials pin only until DiffEqBase registers. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: ChrisRackauckas-Claude <accounts@chrisrackauckas.com> Co-authored-by: Cursor <cursoragent@cursor.com>

* Temporarily source PolyChaos and QMC masters for 32-bit CI. Co-authored-by: Cursor <cursoragent@cursor.com> * Drop QMC git source now that 0.4.3 is registered. Co-authored-by: Cursor <cursoragent@cursor.com> * CI: use SciML/.github#137 so LTS can Pkg.add URL [sources]. Co-authored-by: Cursor <cursoragent@cursor.com> * Require PolyChaos 2.1.2; drop temporary [sources] and #137 pin. 2.1.2 is on master (#193); registration pending. Cross-repo PRs cannot use refs/pull/137/head as a reusable workflow ref. Co-authored-by: Cursor <cursoragent@cursor.com> * Fix PolyChaos UUID accidentally overwritten by compat bump. Co-authored-by: Cursor <cursoragent@cursor.com> * chore: retest after PolyChaos 2.1.2 registration. Co-authored-by: Cursor <cursoragent@cursor.com> * Use factorial(Int64(n)) in Lobachevsky kernels for 32-bit. On i686, Int is Int32 and factorial(n) overflows the Int32 table for n≥13 (tests use n up to 20). Co-authored-by: Cursor <cursoragent@cursor.com> * CI: retrigger Tests after Lobachevsky Int64 factorial fix. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: ChrisRackauckas-Claude <accounts@chrisrackauckas.com> Co-authored-by: Cursor <cursoragent@cursor.com>

* Temporarily source Sundials 32-bit NVector fix for CI. Co-authored-by: Cursor <cursoragent@cursor.com> * CI: use SciML/.github#137 so LTS can Pkg.add URL [sources]. Co-authored-by: Cursor <cursoragent@cursor.com> * Revert #137 workflow pin; drop temporary Sundials [sources]. Await Sundials#567 release / SciML/.github#137+v1 retag for LTS URL sources. Co-authored-by: Cursor <cursoragent@cursor.com> * Re-source Sundials 32-bit NVector fix; CI on SciML/.github@master. Co-authored-by: Cursor <cursoragent@cursor.com> * Retest against Sundials tip without DiffEqBase [sources]. Co-authored-by: Cursor <cursoragent@cursor.com> * Also source DiffEqBase 32-bit interp_points fix. Pin DiffEqBase directly (ODE#4487) alongside Sundials so nested [sources] subdir resolution is not required on julia-pre / i686. Co-authored-by: Cursor <cursoragent@cursor.com> * Declare DiffEqBase test extra so [sources] pin applies. Co-authored-by: Cursor <cursoragent@cursor.com> * Drop DiffEqBase [sources]; keep Sundials pin only. Co-authored-by: Cursor <cursoragent@cursor.com> * Drop x86 CI lane (SBML.jl i686 precompile). SBML fails to precompile on i686 with ArgumentError converting NULL to string during libsbml init. Remove temporary Sundials [sources]. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: ChrisRackauckas-Claude <accounts@chrisrackauckas.com> Co-authored-by: Cursor <cursoragent@cursor.com>

ChrisRackauckas-Claude · 2026-09-10T08:52:27Z

Still fully green and blocking i686 URL [sources] clones (gitconfig / EEXISTS). Please merge + retag floating v1 when you can.

Needed so MCI URL [sources] clones work on x86 before SciML/.github#137 merges to master. Co-authored-by: Cursor <cursoragent@cursor.com>

ChrisRackauckas-Claude · 2026-09-10T09:02:18Z

Note for consumers while this waits on merge: grouped-tests.yml still calls SciML/.github/.../tests.yml@master, so the i686 gitconfig fix in this PR's tests.yml is not exercised via @master until merge.

Temporary consumable tip (self-refs for tests.yml):
ChrisRackauckas-Claude/.github/.github/workflows/grouped-tests.yml@i686-gitconfig-consumable

Integrals#… / BVDE#634 / MomentClosure#118 point at that tip for now; they should revert to SciML/.github@master once this merges.

ChrisRackauckas-Claude · 2026-09-10T09:12:58Z

Fully green and mergeable. Please merge when convenient and retag floating v1 (and cut v1.29.0 if that is the release process).

Several packages currently pin ChrisRackauckas-Claude/.github@i686-gitconfig-consumable for URL [sources] on i686 (Integrals, BVDE, MTK, MomentClosure, StructuralIdentifiability). After this merges + v1 is retagged, those pins can switch back to SciML/.github@master / @v1.

Recreate via tee + chmod, verify with ls, and set GIT_CONFIG_NOSYSTEM=1. Bare touch was insufficient on some ubuntu-latest x86 runners. Co-authored-by: Cursor <cursoragent@cursor.com>

/etc/gitconfig alone still failed on SELinux runners (mode bits with a trailing '.'): recreate with restorecon/chcon, and point LibGit2's SYSTEM search path at $HOME via startup.jl before any URL [sources] clone. Co-authored-by: Cursor <cursoragent@cursor.com>

ChrisRackauckas-Claude · 2026-09-10T10:02:31Z

Pushed follow-up: SELinux restorecon/chcon on /etc/gitconfig plus LibGit2 SYSTEM search-path redirect via ~/.julia/config/startup.jl — MTK x86 still hit failed to stat '/etc/gitconfig' on a runner whose ls -la showed SELinux mode bits (-rw-r--r--.`) even after the file existed.

…ults. On some i686 runners LibGit2 HTTPS clones crash in SSL_CTX_load_verify_file even with /etc/gitconfig fixed. Clone via amd64 git and rewrite url→path. Co-authored-by: Cursor <cursoragent@cursor.com>

ChrisRackauckas-Claude · 2026-09-10T10:53:45Z

Follow-up: i686 LibGit2 was still segfaulting in SSL_CTX_load_verify_file when cloning URL [sources] (Integrals/MTK), even with SELinux-fixed /etc/gitconfig.

Consumable tip now prefetches URL [sources] with host (amd64) git, rewrites them to path = in the runner workspace, and checks out scripts from this tip (not SciML master). After merge, the script checkout should switch back to SciML/.github + dotgithub-ref.

@ChrisRackauckas

* Accept Integer (incl. Int64) dict values in power-series helpers. On 32-bit Julia, Int is Int32 so Dict{P,Int64} from literal keys missed the Dict{P,Int} methods and errored in SI tests. Co-authored-by: Cursor <cursoragent@cursor.com> * Accept empty Dict() inputs in Integer power-series helpers. Call sites pass Dict() for no inputs (Dict{Any,Any}); requiring Dict{P,Array{T,1}} missed that case on both 32- and 64-bit. Co-authored-by: Cursor <cursoragent@cursor.com> * Accept empty Dict() inputs in FieldElem power-series helpers too. FieldElem callers also pass Dict() for no inputs; only the Integer convenience path had been widened. Co-authored-by: Cursor <cursoragent@cursor.com> * Temporarily source Groebner CRT UInt fix for 32-bit precompile. Co-authored-by: Cursor <cursoragent@cursor.com> * CI: use SciML/.github#137 so LTS can Pkg.add URL [sources]. Floating v1 still has develop_sources that Pkg.develop-s rev pins. Co-authored-by: Cursor <cursoragent@cursor.com> * Revert #137 workflow pin; drop temporary Groebner [sources]. Cross-repo reusable workflows cannot use refs/pull/*/head. LTS develop_sources still hits stale v1 until SciML/.github#137 merges and v1 is retagged; Integer/Dict fixes remain. Co-authored-by: Cursor <cursoragent@cursor.com> * Require Groebner 0.10.8 for 32-bit CRT UInt64 fix. Co-authored-by: Cursor <cursoragent@cursor.com> * Use factorial(big(...)) in differentiate_output tests on 32-bit. prec=20 hits factorial(13)+, which overflows Int32's factorial table. Co-authored-by: Cursor <cursoragent@cursor.com> * Source Nemo 32-bit fpMatrix fix; CI scripts from SciML/.github master. Nemo's Julia-owned nmod matrix layout raises InexactError on i686 when storing heap pointers in Vector{Int32}. Temporarily pin the upstream fix branch until it is registered. Co-authored-by: Cursor <cursoragent@cursor.com> * CI: retest now that Nemo fork branch is published. Co-authored-by: Cursor <cursoragent@cursor.com> * Skip Nemo mod-p differentiate_output cases on 32-bit. Drop temporary Nemo [sources] (i686 LibGit2 clone flakes). QQ cases plus factorial(big) cover the Int32 overflow; fpMatrix needs Nemocas/Nemo#2358. Co-authored-by: Cursor <cursoragent@cursor.com> * Skip primality zerodim tests on 32-bit until Nemo#2358. check_primality_zerodim hits Nemo finite-field matrix construction that errors on i686; keep the suite green without masking x64 coverage. Co-authored-by: Cursor <cursoragent@cursor.com> * Fix typos false positive in primality skip comment. Co-authored-by: Cursor <cursoragent@cursor.com> * Wrap all Nemo mod-p differentiate_output cases in WORD_SIZE==64. Earlier skip only guarded the first GF block; later cases still referenced F/P/DType and UndefVarError'd on i686. Co-authored-by: Cursor <cursoragent@cursor.com> * Re-source Nemo#2358 with i686-gitconfig consumable CI tip. URL [sources] previously flaked on i686 LibGit2; pin grouped-tests to the consumable tip that creates /etc/gitconfig and clears partial clones. Disable Downgrade while URL sources are present. Co-authored-by: Cursor <cursoragent@cursor.com> * Skip Nemo fpMatrix identifiability bodies on 32-bit; drop Nemo sources. URL-sourcing Nemo#2358 still fails i686 LibGit2 gitconfig clones even after touching /etc/gitconfig. Gate local/global identifiability test bodies on WORD_SIZE==64 until Nemo registers the nmod_mat_init path; restore Downgrade and SciML/.github@master CI. Co-authored-by: Cursor <cursoragent@cursor.com> * Update linear_compartment_model doctest for Dict iteration order. Integer-keyed graph dicts no longer print ODE equations in the old insertion order; refresh the jldoctest expected output. Co-authored-by: Cursor <cursoragent@cursor.com> * Gate Nemo fpMatrix tests at the SafeTestsets include. Top-level `return` inside included body files does not stop SafeTestsets from evaluating the rest of the file, so 32-bit still hit InexactError. Skip the include itself when Sys.WORD_SIZE != 64. Co-authored-by: Cursor <cursoragent@cursor.com> * Apply suggestion from @ChrisRackauckas --------- Co-authored-by: ChrisRackauckas-Claude <accounts@chrisrackauckas.com> Co-authored-by: Cursor <cursoragent@cursor.com>

ChrisRackauckas-Claude pushed a commit to ChrisRackauckas-Claude/Surrogates.jl that referenced this pull request Sep 9, 2026

CI: use SciML/.github#137 so LTS can Pkg.add URL [sources].

0f82f04

Co-authored-by: Cursor <cursoragent@cursor.com>

ChrisRackauckas-Claude pushed a commit to ChrisRackauckas-Claude/SteadyStateDiffEq.jl that referenced this pull request Sep 9, 2026

CI: use SciML/.github#137 so LTS can Pkg.add URL [sources].

cb17c0a

Co-authored-by: Cursor <cursoragent@cursor.com>

ChrisRackauckas-Claude pushed a commit to ChrisRackauckas-Claude/SBMLToolkit.jl that referenced this pull request Sep 9, 2026

CI: use SciML/.github#137 so LTS can Pkg.add URL [sources].

0d82d6f

Co-authored-by: Cursor <cursoragent@cursor.com>

ChrisRackauckas and others added 2 commits September 9, 2026 08:08

Point grouped-tests at this branch's tests.yml (not floating v1).

e856ff4

Otherwise the i686 /etc/gitconfig + clones cleanup in tests.yml is never executed when packages consume this PR tip. Co-authored-by: Cursor <cursoragent@cursor.com>

Use tests.yml@master from grouped-tests (not floating v1).

49ef2b2

After merge, master carries the i686 gitconfig/clones workaround; packages on grouped-tests@master then execute that tests.yml tip. Co-authored-by: Cursor <cursoragent@cursor.com>

ChrisRackauckas-Claude mentioned this pull request Sep 9, 2026

Sundials 6.7.1 + MCI 0.3 source for x86 InterfaceI SciML/ModelingToolkit.jl#5098

Closed

1 task

This was referenced Sep 10, 2026

Source MCI 0.3 for x86 LatticeRules compat SciML/BoundaryValueDiffEq.jl#634

Closed

CI: i686 gitconfig tip for MCI URL [sources] SciML/Integrals.jl#367

Closed

ChrisRackauckas-Claude mentioned this pull request Sep 10, 2026

Accept Int64 dict values in power_series_solution on 32-bit SciML/StructuralIdentifiability.jl#555

Merged

2 tasks

Harden i686 /etc/gitconfig setup for LibGit2 URL clones.

ed1550f

Recreate via tee + chmod, verify with ls, and set GIT_CONFIG_NOSYSTEM=1. Bare touch was insufficient on some ubuntu-latest x86 runners. Co-authored-by: Cursor <cursoragent@cursor.com>

ChrisRackauckas-Claude requested a review from ChrisRackauckas September 10, 2026 09:40

i686: prefetch URL [sources] with host git to avoid LibGit2 SSL segfa…

2da656a

…ults. On some i686 runners LibGit2 HTTPS clones crash in SSL_CTX_load_verify_file even with /etc/gitconfig fixed. Clone via amd64 git and rewrite url→path. Co-authored-by: Cursor <cursoragent@cursor.com>

ChrisRackauckas closed this Sep 11, 2026

Navigation Menu

Forward dotgithub-ref to tests.yml; default to master - #137

Forward dotgithub-ref to tests.yml; default to master#137
ChrisRackauckas-Claude wants to merge 7 commits into
SciML:masterfrom
ChrisRackauckas-Claude:fix-forward-dotgithub-ref-default-master

ChrisRackauckas-Claude commented Sep 9, 2026

Uh oh!

ChrisRackauckas-Claude commented Sep 9, 2026

Uh oh!

ChrisRackauckas-Claude commented Sep 9, 2026

Uh oh!

ChrisRackauckas-Claude commented Sep 9, 2026

Uh oh!

ChrisRackauckas-Claude commented Sep 9, 2026

Uh oh!

ChrisRackauckas-Claude commented Sep 9, 2026

Uh oh!

ChrisRackauckas-Claude commented Sep 9, 2026

Uh oh!

ChrisRackauckas-Claude commented Sep 10, 2026

Uh oh!

ChrisRackauckas-Claude commented Sep 10, 2026

Uh oh!

ChrisRackauckas-Claude commented Sep 10, 2026

Uh oh!

ChrisRackauckas-Claude commented Sep 10, 2026

Uh oh!

ChrisRackauckas-Claude commented Sep 10, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

ChrisRackauckas-Claude commented Sep 9, 2026

Summary

Test plan

Uh oh!

ChrisRackauckas-Claude commented Sep 9, 2026

Uh oh!

ChrisRackauckas-Claude commented Sep 9, 2026

Uh oh!

ChrisRackauckas-Claude commented Sep 9, 2026

Uh oh!

ChrisRackauckas-Claude commented Sep 9, 2026

Merge request (32-bit SciML CI parity)

Uh oh!

ChrisRackauckas-Claude commented Sep 9, 2026

Merge request (32-bit SciML CI parity) — elevated

Uh oh!

ChrisRackauckas-Claude commented Sep 9, 2026

Uh oh!

ChrisRackauckas-Claude commented Sep 10, 2026

Uh oh!

ChrisRackauckas-Claude commented Sep 10, 2026

Uh oh!

ChrisRackauckas-Claude commented Sep 10, 2026

Uh oh!

ChrisRackauckas-Claude commented Sep 10, 2026

Uh oh!

ChrisRackauckas-Claude commented Sep 10, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants