[Lldb-commits] [PATCH] D77043: Fix process gdb-remote usage of value_regs/invalidate_regs

Pavel Labath via Phabricator via lldb-commits lldb-commits at lists.llvm.org
Tue Apr 14 06:22:49 PDT 2020

labath added a comment.

In D77043#1980208 <https://reviews.llvm.org/D77043#1980208>, @omjavaid wrote:

> For current implementation I dont think it will break any stubs because newly introduced regnum is totally optional. If regnum is not provided then mocked up register index is used and things will work as they were.

I believe it is optional right now, but that can easily change in the future. That's why I am looking into whether it's possible to make this work without introducing the new field. If all code takes the same path, its much less likely something will break.

> On a different note LLDB now supports sending over target xml packet and there it has to send a register number along with register name and everything else. Most of stubs like QEMU, OpenOCD, etc use target xml for register description exchange and we may consider giving up qRegisterInfo in favor of target xml in future.

Lldb is able to use target.xml too. However, relying on as the sole source of register numbers would effectively make libxml a mandatory dependency. Maybe that would be possible, but that's a topic for a wider discussion, as some platforms (windows) don't have libxml easily available. Anyway, falling back to a different format is not what worries me that much -- I'm more concerned if we diverge on the content we send using those formats from what the other stubs send.



More information about the lldb-commits mailing list