Re: MPS image cache

[Gerd Möllmann]

Eli Zaretskii <eliz@gnu.org> writes:

>> From: Gerd Möllmann <gerd.moellmann@gmail.com>
>> Cc: eller.helmut@gmail.com,  emacs-devel@gnu.org
>> Date: Sun, 05 May 2024 14:55:04 +0200
>> 
>> Eli Zaretskii <eliz@gnu.org> writes:
>> 
>> >> It is on my platform.
>> >
>> > IME, this cannot be relied upon, unless we explicitly tell the
>> > compiler to pack (using some pragma or somesuch).
>> 
>> And the static_assert will fire, right. You're telling me nothing new,
>> BTW. I'm pretty well acquainged with all this.
>
> I imagined, it just took me by surprise that you would nonetheless
> write code that could so easily fail to compile.

I think it's a matter of personal priorities. It guess it can't hurt to
say explicity what these are.

1. Make progress fast, so that I can assess for myself if it's worth
continuing. With an Emacs up time of ca. 8h now, I tend to think so.
When it would reach say 5 days or so, we'd be in the realm of what
master survives for me on macOS.

2. Make it work for me in my personal fork (CL packages) which is macOS
only anyway, and is basically for me alone only.

3. GNU. And I'm neither a maintainer of anything, nor want to maintain
anything in the GNU sense.

>> >> We allocate a vector of N pointers. The N is found in the header when we
>> >> scan it. It's the size of the MPS object we scan minus the header
>> >> divided by how large a pointer is.
>> >
>> > OK, but why does the struct have to fit in a single word?
>> 
>> It doesn't, it the other way around. I'm using nwords - 1 for the size
>> of the vector, and the static_assert is intended to fire if nwords - 1
>> is not right.
>
> OK, then my question now is "why does it have to be nwords-1?"

I suspect that I'll answer something you don't ask again: We allocate a
vector of N elements. We don't store N directly somewhere, but we have
an igc_header from which we can compute N. So I compute it as nwords -
1. If that's not right, I static_assert, and leave it to the interested
party to do something.