Re: [PATCH v2 1/1] Make __sdidinit unused

Hi Seb,

On Feb 18 17:21, Sebastian Huber wrote:
> Hello Corinna,
> 
> On 18/02/2022 13:41, Corinna Vinschen wrote:
> > Hi Matt,
> > 
> > On Feb 18 10:45, Matthew Joyce wrote:
> > > Remove dependency on __sdidinit member of struct _reent to check
> > > object initialization. Like __sdidinit, the __cleanup member of
> > > struct _reent is initialized in the __sinit() function. Checking
> > > initialization against __cleanup serves the same purpose and will
> > > reduce overhead in the __sfp() function in a follow up patch.
> > 
> > The patch looks right now.  But.
> > 
> > What exactly are you going to do in __sfp?  The reason I'm asking is
> > that it's absolutely not clear yet which purpose this change serves.
> > Looking at this patch, I only see that, rather than using an available
> > flag, a pointer is now set to an invalid value -1, which looks more
> > dangerous than what we did before.
> > 
> > Care to explain or, even better, send the patch you have in mind?
> 
> this is related to our attempt to use individual thread-local storage
> objects instead of the monolithic struct _reent:
> 
> https://sourceware.org/pipermail/newlib/2022/018855.html
> 
> It turned out that this is not possible while _GLOBAL_REENT exists. Getting
> rid of struct _reent members with similar functionality helps to refactor
> the code.
> 
> Instead of using -1 as a magic value a safer option would be to assign a
> dummy function which does nothing:
> [...]

I like the idea.  This looks good in the code.  A comment in
_cygtls::init_thread might be helpful in future, but it's not
required for this patch submission.

Matt, can you pleae resubmit with this change?


Thanks,
Corinna