Date: Tue, 4 Jan 2022 21:16:23 -0500
From: Christopher Wellons <wellons@nullprogram.com>
To: Eric Farmer <erfarmer@gmail.com>
Cc: ~skeeto/public-inbox@lists.sr.ht
Subject: Re: A new protocol and tool for PNG file attachments
Message-ID: <20220105021623.6wq5zlciqfgtdlrp@nullprogram.com>
References: <CAAuXPZoLFAn1o-c1+yKrVL1qQtBS5U_WWW5UJseUJjjgYXWLCw@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <CAAuXPZoLFAn1o-c1+yKrVL1qQtBS5U_WWW5UJseUJjjgYXWLCw@mail.gmail.com>
User-Agent: NeoMutt/20170113 (1.7.2)

Thanks, Eric! That very project being shared around was what got the gears 
turning for my project. I was inspired by the headline, skimmed the PNG 
specification to brush up on the format, went to sleep, and by morning I 
had a few ideas for how to tackle it. I Initially considered compatibility 
with pngsource (i.e. the draw.io format), but saw the XML and noped out.

> what motivates the `void *` buffer type in `crc32_update`

That's just the result of a copy-paste incompletely integrated into the 
new project. :-)

https://github.com/skeeto/scratch/blob/master/checksums/crc32.h

The original is C89 — no stdint.h, no for loop variable declaration — and, 
being agnostic of the surrounding context, accepts a generic block of 
memory, which as you pointed out could be a struct. I like to avoid even 
"char *" when accepting raw memory since it's opinionated about the sign 
of its input.

As I worked, I massaged it into the context of pngattach, adapting it to 
C99 and fixed-width integers. Since I use "unsigned char" consistently 
throughout the rest of the program as an alias for raw PNG bytes, I 
probably should have changed that aspect of its interface as well.

Side note: In the past when processing string data ("char *") as though 
unsigned bytes, such as to decode UTF-8 or do table lookups, I've cast 
"char *" to "unsigned char *". However, I now use "&0xff" on the read 
value to force it to the equivalent unsigned value without the awkward 
pointer cast. Every compiler I've tested handles this perfectly (turns it 
into a zero-extension), so there's no penalty. Technically this requires a 
two's complement representation, but practically speaking, none of my 
programs will ever see a computer that isn't so.