step-avoid-regexp not working

[ilev4ik]

Hi! I cannot figure out a reason of not working of 'step-avoid-regexp' for my executable, so created this issue.

Preconditions:

lldb version 16.0.6
python 3.10 x64
LLDB_USE_NATIVE_PDB_READER=1

If LLDB_USE_NATIVE_PDB_READER=1 is not set lldb crashes on, for example, "fr v".

I unsuccessfully tried to find out why msdia140.dll fails by procmon. So decided to switch to native pdb reader.

I compile with msvc and use /Ob1 to avoid inlining, and doublechecked source files are correctly opened when reached a breakpoint.

Typical compilation command (extra options removed)
cl.exe /MDd /external:W0 /external:templates- /DUNICODE /D_UNICODE /Od /c path/to/source/file.cpp /nologo /Z7 /bigobj /Zc:forScope /Zc:inline /Zc:wchar_t /fp:precise /D_CRT_NONSTDC_NO_WARNINGS /D_CRT_SECURE_NO_DEPRECATE /D_CRT_SECURE_NO_WARNINGS /D_SCL_SECURE_NO_WARNINGS /DWIN32 /D_WINDOWS /D_WINSOCK_DEPRECATED_NO_WARNINGS /DWINVER=0x0600 /D_WIN32_WINNT=0x0600 /DNTDDI_VERSION=0x06000000 /W3 /guard:cf /GR- /EHsc /guard:ehcont /source-charset:windows-1251 /execution-charset:windows-1251 /Oy- **/Ob1** /WX /wd4996 /wd4117 -D__DATE__="redacted" /RTC1

Launching and configuration:

lldb.exe proc.exe
> settings set target.source-map . F:/prj/mono/src
> settings set target.process.thread.step-avoid-regexp ^.*(std::|boost::).*$
> b main
> r
> si

But anyway step-into falls to std:: and boost:: functions which I expect should not happen.

Even more: if I set regexp=.* nothing changed.

• Is this behaviour expected?
• May this happen because of preconditions (especially LLDB_USE_NATIVE_PDB_READER=1)?
• Any recommendation or workarounds?

TY

[EugeneZelenko]

Could you please try 20 or main branch?

[llvmbot]

@llvm/issue-subscribers-lldb

Author: Oganesyan Levon (ilev4ik)

Hi! I cannot figure out a reason of not working of 'step-avoid-regexp' for my executable, so created this issue.

Preconditions:

lldb version 16.0.6
python 3.10 x64
LLDB_USE_NATIVE_PDB_READER=1

If LLDB_USE_NATIVE_PDB_READER=1 is not set lldb crashes on, for example, "fr v".
> I unsuccessfully tried to find out why msdia140.dll fails by procmon. So decided to switch to native pdb reader.

I compile with msvc and use /Ob1 to avoid inlining, and doublechecked source files are correctly opened when reached a breakpoint.

Typical compilation command (extra options removed)
cl.exe /MDd /external:W0 /external:templates- /DUNICODE /D_UNICODE /Od /c path/to/source/file.cpp /nologo /Z7 /bigobj /Zc:forScope /Zc:inline /Zc:wchar_t /fp:precise /D_CRT_NONSTDC_NO_WARNINGS /D_CRT_SECURE_NO_DEPRECATE /D_CRT_SECURE_NO_WARNINGS /D_SCL_SECURE_NO_WARNINGS /DWIN32 /D_WINDOWS /D_WINSOCK_DEPRECATED_NO_WARNINGS /DWINVER=0x0600 /D_WIN32_WINNT=0x0600 /DNTDDI_VERSION=0x06000000 /W3 /guard:cf /GR- /EHsc /guard:ehcont /source-charset:windows-1251 /execution-charset:windows-1251 /Oy- **/Ob1** /WX /wd4996 /wd4117 -D__DATE__="redacted" /RTC1

Launching and configuration:

lldb.exe proc.exe
> settings set target.source-map . F:/prj/mono/src
> settings set target.process.thread.step-avoid-regexp ^.*(std::|boost::).*$
> b main
> r
> si

But anyway step-into falls to std:: and boost:: functions which I expect should not happen.

Even more: if I set regexp=.* nothing changed.

• Is this behaviour expected?
• May this happen because of preconditions (especially LLDB_USE_NATIVE_PDB_READER=1)?
• Any recommendation or workarounds?

TY

[jimingham]

The command you are using: si does not mean "step into" it means "step-instruction". The lldb command that does "step into" is just step.

step-instruction isn't restricted by the "step-avoid-regex" - its guarantee is that it will always only step one instruction and then stop, so it can't do step outs and so forth.

[ilev4ik]

@jimingham I checked - step doesn't work also

I faced the same issue when working from vscode's lldb extension - f11 works as I described above

[jimingham]

I don't know how f11 in VSCode implements its step. I tried your regex with the current TOT lldb and running the step command in the lldb command interpreter, and it correctly avoided stepping into inlined std code, but that was on macOS not Windows?

This is a pretty simple facility, it just compares that regex against the demangled name of the current function, and if it matches, we step back out again. So either the regex is failing, or we can't figure out how to step out of the function.

You can check the regex part pretty easily, get the name of the function you shouldn't have stopped in and see if your regex matches it.

If that did match, maybe we can't step out. That should be straightforward since this is inlined code so we should just step till we are no longer in the inlining section, but maybe the line table is funny and we can't figure out how to do that.

The next step to actually debug this is to get lldb to emit the step log by running:

(lldb) log enable -f /tmp/lldb-log.txt lldb step

then doing the "step" command over some std:: function. That will show you the stages of this process.