Em sáb., 5 de set. de 2020 às 14:29, Tom Lane <tgl@sss.pgh.pa.us> escreveu:
Ranier Vilela <ranier.vf@gmail.com> writes: > Attached is a patch I made in March/2020, but due to problems, > it was sent but did not make the list. > Would you mind taking a look?
I applied some of this, but other parts had been overtaken by events, and there were other changes that I didn't agree with.
I fully agree with your judgment.
A general comment on the sort of "dead store" that I don't think we should remove is where a function is trying to maintain an internal invariant, such as "this pointer points past the last data written to a buffer" or "these two variables are in sync". If the update happens to be the last one in the function, the compiler may be able to see that the store is dead ... but IMO it should just optimize such a store away and not get in the programmer's face about it. If we manually remove the dead store then what we've done is broken the invariant, and we'll pay for that in future bugs and maintenance costs. Somebody may someday want to add more code after the step in question, and if they fail to undo the manual optimization then they've got a bug. Besides which, it's confusing when a function does something the same way N-1 times and then differently the
N'th time.
Good point. The last store is a little strange, but the compiler will certainly optimize. Maintenance is expensive, and the current code should be the best example.