[Bug record/24072] New: Stepping state is not cleared on out-of-history stop

["autkin at undo dot io" <sourceware-bugzilla at sourceware dot org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=24072

            Bug ID: 24072
           Summary: Stepping state is not cleared on out-of-history stop
           Product: gdb
           Version: HEAD
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: record
          Assignee: unassigned at sourceware dot org
          Reporter: autkin at undo dot io
  Target Milestone: ---

Created attachment 11514
  --> https://sourceware.org/bugzilla/attachment.cgi?id=11514&action=edit
stress test script

The lack of clear_step_over_info() call in TARGET_WAITKIND_NO_HISTORY handling
causes troubles in a sequence of stepping and record-goto commands.

With GDB 8.1 in conjunction with Undo reversible debugging server, we have hit
a "No unwaited children left" condition after something like

* go to start of recorded history;
* reverse-stepi;
* go to start of recorded history.

GDB's native reversible-debugging interface won't hit this because it would
report "Already at target insn" and won't do anything. However, the following
stress-test hits this bug in GDB by hitting the beginning and the end of the
recording extent, which are exactly one instruction apart. The stress-test
causes GDB to hang. I have not investigated exact mechanics of why it hangs,
sorry, but I do confirm that the supplied one-liner patch fixes it.

Speaking of the fixing patch, I would ask maintainers to consider another
patch, which still gets this bug fixed:
https://sourceware.org/ml/gdb-patches/2019-01/msg00074.html . The patch makes
more changes, I would be grateful for discussion  of it.

-- 
You are receiving this mail because:
You are on the CC list for the bug.