[Lldb-commits] [PATCH] D53761: [Target] Do not skip a stop on a breakpoint if a plan was completed

Aleksandr Urakov via Phabricator via lldb-commits lldb-commits at lists.llvm.org
Fri Oct 26 07:11:16 PDT 2018

aleksandr.urakov created this revision.
aleksandr.urakov added reviewers: jingham, zturner, boris.ulasevich.
aleksandr.urakov added a project: LLDB.
Herald added subscribers: lldb-commits, teemperor.

This patch fixes the next situation. On Windows `clang-cl` makes no stub before the `main` function, so the `main` function is located exactly on module entry point. May be it is the same on other platforms. So consider the following sequence:

- set a breakpoint on `main` and stop there;
- try to evaluate expression, which requires a code execution on the debuggee side. Such an execution always returns to the module entry, and the plan waits for it there;
- the plan understands that it is complete now and removes its breakpoint. But the breakpoint site is still there, because we also have a breakpoint on entry;
- `StopInfo` analyzes a situation. It sees that we have stopped on the breakpoint site, and it sees that the breakpoint site has owners, and no one logical breakpoint is internal (because the plan is already completed and it have removed its breakpoint);
- `StopInfo` thinks that it's a user breakpoint and skips it to avoid recursive computations;
- the program continues.

So in this situation the program continues without a stop right after the expression evaluation. To avoid this I've added an additional check that the plan was completed.

The test is just the case I've caught, but it's Windows-only. It seems that the problem should exist on other platforms too, but I don't know how to test it in a cross-platform way (any tips are appreciated). The test depends on https://reviews.llvm.org/D53759.




Index: source/Target/StopInfo.cpp
--- source/Target/StopInfo.cpp
+++ source/Target/StopInfo.cpp
@@ -333,6 +333,13 @@
             // commands when we see the same breakpoint hit a second time.
             m_should_stop_is_valid = true;
+            if (thread_sp->CompletedPlanOverridesBreakpoint()) {
+              m_should_stop = true;
+              thread_sp->ResetStopInfo();
+              return;
+            }
             if (log)
               log->Printf("StopInfoBreakpoint::PerformAction - Hit a "
                           "breakpoint while running an expression,"
Index: lit/SymbolFile/PDB/bp-on-entry.test
--- /dev/null
+++ lit/SymbolFile/PDB/bp-on-entry.test
@@ -0,0 +1,7 @@
+REQUIRES: windows
+RUN: clang-cl /Zi /GS- /c %S/Inputs/BPOnEntryTest.cpp /Fo%t.obj
+RUN: lld-link /debug:full /nodefaultlib /entry:main %t.obj /out:%t.exe
+RUN: %lldb -b -s %S/Inputs/BPOnEntryTest.script -- %t.exe | FileCheck %s
+CHECK: (lldb) print sum(3, 4)
+CHECK: (int) $0 = 7
Index: lit/SymbolFile/PDB/Inputs/BPOnEntryTest.script
--- /dev/null
+++ lit/SymbolFile/PDB/Inputs/BPOnEntryTest.script
@@ -0,0 +1,3 @@
+breakpoint set --file BPOnEntryTest.cpp --name main
+print sum(3, 4)
Index: lit/SymbolFile/PDB/Inputs/BPOnEntryTest.cpp
--- /dev/null
+++ lit/SymbolFile/PDB/Inputs/BPOnEntryTest.cpp
@@ -0,0 +1,7 @@
+int sum(int a, int b) {
+  return a + b;
+int main() {
+  return sum(-1, 1);

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D53761.171296.patch
Type: text/x-patch
Size: 1684 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/lldb-commits/attachments/20181026/6c66e028/attachment.bin>

More information about the lldb-commits mailing list