webkit2gtk3/revert-271175.patch
Dominique Leuenberger 15cddb5c5d - Update to version 2.44.4:
+ Add quirk to allow totale.rosettastone.com to load properly.
  + Fix webkit_web_resource_get_data() not working properly in some
    sites.
  + Fix not being able to jump-to-source in Web Inspector canvas
    traces.
  + Fix not being able to scroll list of WebGL shader programs in
    the Web Inspector.
  + Fix linker relocation errors on Debug/RelWithDebInfo builds.
  + Fix crashes when built with Clang with Link-Time Optimization
    (LTO).
  + Fix several crashes and rendering issues.
- Drop revert-271175.patch: Fixed upstream.
- Enable LTO again, pass define _lto_cflags for only select targets

OBS-URL: https://build.opensuse.org/package/show/GNOME:Factory/webkit2gtk3?expand=0&rev=479
2024-09-11 14:28:56 +00:00

48 lines
2.6 KiB
Diff

From 9140ce712aa87091613874d802787ab476be0e39 Mon Sep 17 00:00:00 2001
From: Michael Catanzaro <mcatanzaro@redhat.com>
Date: Wed, 14 Aug 2024 14:58:05 -0500
Subject: [PATCH] Revert "Cherry-pick 272448.770@safari-7618-branch
(6d311cd7fefc). https://bugs.webkit.org/show_bug.cgi?id=271175"
https://bugs.webkit.org/show_bug.cgi?id=278113
This reverts commit 279c9d7963182cc35cf4e0bfebe87df2d83eaef8.
This broke wasm, and I don't know how to fix it.
Canonical link: https://commits.webkit.org/274313.373@webkitglib/2.44
---
.../stress/many-calls-results-on-stack.js | 39 -------------------
Source/JavaScriptCore/wasm/WasmBBQJIT.cpp | 19 ---------
2 files changed, 58 deletions(-)
delete mode 100644 JSTests/wasm/stress/many-calls-results-on-stack.js
index 9049865e8ce7..3f142cf5e90f 100644
--- a/Source/JavaScriptCore/wasm/WasmBBQJIT.cpp
+++ b/Source/JavaScriptCore/wasm/WasmBBQJIT.cpp
@@ -3958,25 +3958,6 @@ void BBQJIT::returnValuesFromCall(Vector<Value, N>& results, const FunctionSigna
ASSERT(m_validFPRs.contains(returnLocation.asFPR(), Width::Width128));
m_fprSet.add(returnLocation.asFPR(), Width::Width128);
}
- } else {
- ASSERT(returnLocation.isStackArgument());
- // FIXME: Ideally, we would leave these values where they are but a subsequent call could clobber them before they are used.
- // That said, stack results are very rare so this isn't too painful.
- // Even if we did leave them where they are, we'd need to flush them to their canonical location at the next branch otherwise
- // we could have something like (assume no result regs for simplicity):
- // call (result i32 i32) $foo
- // if (result i32) // Stack: i32(StackArgument:8) i32(StackArgument:0)
- // // Stack: i32(StackArgument:8)
- // else
- // call (result i32 i32) $bar // Stack: i32(StackArgument:8) we have to flush the stack argument to make room for the result of bar
- // drop // Stack: i32(Stack:X) i32(StackArgument:8) i32(StackArgument:0)
- // drop // Stack: i32(Stack:X) i32(StackArgument:8)
- // end
- // return // Stack i32(*Conflicting locations*)
-
- Location canonicalLocation = canonicalSlot(result);
- emitMoveMemory(result.type(), returnLocation, canonicalLocation);
- returnLocation = canonicalLocation;
}
}
bind(result, returnLocation);