DO NOT MERGE: test a MultiSSL libcurl (curl[openssl]) on Windows arm64 - #6755
Closed
Fedr wants to merge 26 commits into
Closed
DO NOT MERGE: test a MultiSSL libcurl (curl[openssl]) on Windows arm64#6755Fedr wants to merge 26 commits into
Fedr wants to merge 26 commits into
Conversation
Adds curl[openssl] to the Windows vcpkg requirements, which makes vcpkg build libcurl with both OpenSSL and Schannel (MultiSSL) instead of the Schannel-only default, and trims the Windows minimal matrix to the two arm64 legs so only that configuration is exercised. Purpose: a downstream Windows app that requests curl[openssl] crashes with an access violation inside cpr::Session::Get on the first HTTPS request on arm64 only, while MeshLib's Schannel-only arm64 build passes MRViewer.CPRTestGet/CPRTestPost. This branch checks whether the crash follows the MultiSSL libcurl into MeshLib. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Fedr
added
skip-image-rebuild
Adds a raw-libcurl gtest (no cpr, every step logged, curl verbose trace through spdlog) and a verbose cpr session test, then runs each of them in its own process, with and without CURL_SSL_BACKEND=schannel, and copies vcpkg's PDBs next to the binaries so the crash stack resolves to function and line instead of nearest export. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The default handler prints only the faulting data address, and unwinding through OpenSSL's arm64 code fails (no libssl/libcrypto frames appear at all), so the stack has been useless. A vectored handler now reports the faulting PC as module+RVA, every general register, and the module+RVA of every return-address-looking word near SP. Adds a raw-libcurl variant that sets CURLSSLOPT_NATIVE_CA, and runs each probe in its own process. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
vcpkg already ships debug libssl/libcrypto in the same package, so the release/debug question needs no rebuild: back up the release DLLs, drop the debug ones (plus the debug CRT they need) next to the binaries, and rerun the raw libcurl request. If it passes, the defect is optimization-dependent codegen rather than OpenSSL logic. The fault handler now also prints the module's PE timestamp/size and the nine instructions around the faulting PC, so the running code can be matched against an offline disassembly with certainty. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
scripts/arm64_openssl_mre.c is ~90 lines of C against libssl/libcrypto only - no curl, no cpr, no C++: open a TLS 1.3 client connection and report where it dies, printing the faulting instruction as module+RVA and the bogus x0. The CI step compiles it with cl /O2 and runs it twice in the same directory, first against the release OpenSSL DLLs and then against the debug ones, so one log shows both outcomes. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The reproducer needs OpenSSL headers, import libs and DLLs - nothing else - so building MeshLib first (submodules, vcpkg restore, MRBind, the whole CMake build, ~25 min) to reach it was pure waste. The new workflow downloads the prebuilt arm64 OpenSSL package straight from the vcpkg binary cache, compiles the reproducer with cl /O2 and runs it twice.
Contributor
Author
|
Done with this branch: it was only ever a diagnostic, and it answered the question. The arm64 crash is not a MeshLib, cpr, curl or MultiSSL problem. It is an MSVC ARM64 code-generation defect: at Reported, with a ~25-line reproducer:
Nothing here is mergeable - the |
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
1 participant
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.
Throwaway CI experiment — do not merge.
MeshLib's Windows arm64 legs run
MRViewer.CPRTestGet/CPRTestPoston every PR and they pass, because plaincprgets a Schannel-only libcurl:A downstream Windows app appends
curl[openssl]to the same requirements file, which makes vcpkg build a MultiSSL libcurl (OpenSSL/3.6.3 (Schannel), backendsopenssl+schannel). With that build, arm64 crashes with an access violation reading0x879insidecpr::Session::Geton the first HTTPS request; the identical x64 build is fine, and forcing the Schannel backend at runtime does not help — so the suspect is the MultiSSL backend dispatch on arm64 rather than OpenSSL's TLS code.This branch adds
curl[openssl]torequirements/windows.txtto see whether the crash follows the MultiSSL libcurl into MeshLib's own arm64 MRTest.Scope trimming (throwaway, so the leg is cheap):
skip-image-rebuildso therequirements/windows.txtchange does not fire the four-tripletwindows-vcpkg-build-uploadprepare job; the build jobs install their own deps with--write-s3.Expected outcome:
MRViewer.CPRTestGetcrashes/fails on arm64 → confirms MultiSSL libcurl as the cause and makes MeshLib the minimal reproducer. If it passes, something outside libcurl's feature set is at fault.