[Lldb-commits] [PATCH] D120320: [lldb/driver] Fix SIGTSTP handling

[Andreu Carminati via Phabricator via lldb-commits]

andcarminati added inline comments.


================
Comment at: lldb/tools/driver/Driver.cpp:673-674
+static void sigtstp_handler(int signo) {
   if (g_driver != nullptr)
     g_driver->GetDebugger().SaveInputTerminalState();
 
----------------
labath wrote:
> JDevlieghere wrote:
> > andcarminati wrote:
> > > JDevlieghere wrote:
> > > > labath wrote:
> > > > > JDevlieghere wrote:
> > > > > > I see an opportunity for a little RAII helper.
> > > > > What kind of a helper did you have in mind? Practically the entire function consists of setup and teardown in preparation for the `raise(signo)` call. If I wanted to be fancy I could put all of that in a helper, but I don't think that would make it cleaner. Plus, we also need to be careful about the functions we call from a signal handler, and I really don't know whether e.g. `llvm::make_scope_exit` is guaranteed to not allocate (heap) memory.
> > > > I was only referring to the Save/RestoreInputTerminalState() part of this function. Something like:
> > > > 
> > > > ```
> > > > class TerminalStateRAII() {
> > > > public:
> > > >   TerminalStateRAII(Driver* driver) : driver(m_driver) {
> > > >     if (m_driver)
> > > >       m_driver->GetDebugger().SaveInputTerminalState();
> > > >   }
> > > > 
> > > >   ~SignalHelper() {
> > > >     if (m_driver)
> > > >       m_driver->GetDebugger().SaveInputTerminalState();
> > > >   }
> > > > 
> > > > private:
> > > >   Driver* m_driver;
> > > > };
> > > > ```
> > > > 
> > > > Obviously, this isn't at all important, just something that came to mind.
> > > I think this is a good idea to reduce code duplication. Another approach:
> > > 
> > > ```
> > > class TerminalStateRAII() {
> > > public:
> > >   TerminalStateRAII(Driver* driver) : driver(m_driver) {
> > >     SaveInputTerminalState();
> > >   }
> > > 
> > >   ~TerminalStateRAII() {
> > >     SaveInputTerminalState();
> > >   }
> > > 
> > > private:
> > >   Driver* m_driver;
> > >   void SaveInputTerminalState(){
> > >     if (m_driver)
> > >       m_driver->GetDebugger().SaveInputTerminalState();
> > >   }
> > > };
> > > 
> > > ```
> > That's a typo on my part, the destructor needs to call `RestoreInputTerminalState` (as opposed to `SaveInputTerminalState`).
> Ok, I see. If this was a more complex function (e.g. multiple return points), then I'd agree, but this function is really simple and linear (just like a signal handler should be). I am not convinced by the "code duplication" argument -- the way I see it, the helper class replaces 4 lines of (simple) code with ~15 lines of boilerplate. And this function is literally the only caller of Save/RestoreInputTerminalState.
Considering this argument:

> Ok, I see. If this was a more complex function (e.g. multiple return points), then I'd agree, but this function is really simple and linear (just like a signal handler should be). 

LGTM


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D120320/new/

https://reviews.llvm.org/D120320