Re: My experiences with Rust

Am 22.08.25 um 19:45 schrieb Poul-Henning Kamp:
> --------
> David Chisnall writes:
> 
>>> Aka: the guy who added kern.geom.confxml in 2002
>>
>> XML is not great for generating or parsing,
> 
> I'll have you know that XML was *THE HOT THING* in 2002 :->
> 
> Also I disagree wrt to "generating", because that's just a matter
> of sbuf_printf() calls, but parsing ?  Yeah, bugger that.
> 

Yeah, and when the structure isn't right you're gonna spend seconds on 
end chewing on a very verbose JSON/XML/... table what used to be "piece 
of cake" in CSV (replace C by tab/semicolon or whatnot).  For structured 
repeatable data, these formats tend to get highly redundant for 
nontrivial amounts of data.

At a different place, we're parsing 250 MB XML stuff, a dozen times 
before our simulation of a real-world system starts.  I haven't looked 
more deeply (session scheduled though), it might have to do with arrays 
being unrolled into XML or something (I might know more a week from 
now), but even without we're likely looking at 30 or 40 MB XML to 
describe something that will end up in a maybe 64 kB binary blob.


Talking about Rust, that supply chain security consideration, complexity 
for writing JSON and the need for formatting stuff... makes me wonder if 
there's more to it than forcing type ownership to be considered more 
thoroughly than in many other non-garbage-collecting languages.

Regarding formatting in C, we have <inttypes.h> and the 
formatting/scanning macros PRI* and SCN* 
<https://en.cppreference.com/w/c/types/integer.html#Types>, and in the 
base system we should have or see to it that we have C++20's <format> 
and C++23's <print>.  And possibly module support to get compile times down.

-- 
Matthias Andree
FreeBSD ports committer