From: Junio C Hamano <gitster@pobox.com>
Date: Sun, 24 Aug 2025 16:11:46 GMT
Subject: Re: [PATCH 0/2] progress: replace setitimer() with alarm()
Message-ID: <xmqqsehgu2bh.fsf@gitster.g>
In-Reply-To: <08f405a6-fd2e-40d7-850a-574356b4009e@kdbg.org>

Johannes Sixt <j6t@kdbg.org> writes:

>> Operating system folks may have worked hard to minimize the cost of
>> system calls to gettimeofday() in order to help applications that do
>> so, but I somehow feel even dirtier to hear proposal to do so to
>> replace a signal that we set and forget, to be reminded once every
>> second.
>
> I think that ship has sailed already. Look at display_throughput(). One
> of the first things it does is to look at the wallclock a.k.a.
> getnanotime().

It can be fixed if we wanted to, though, no?  Instead of doing all
the computation for the latest lap, and then decide not to show by
looking at the progress_update flag (set by the interrupt), we can
accumulate the total in the progress->throughput struct until we see
the progress_update flag, at which time we can look at the wallclock
time, compute the time difference, perform clever division, etc.

> That said, I am not very happy about the new calls introduced in
> display_progress(), either. I'll see whether I can produce some
> performance measurements.
>
> I observe a behavior change with delayed progress indicators that I have
> to understand and fix it before I can submit the cleaned up patches.

Thanks.