Implementation notes

A program is complete when the schedule dictates that it's done.

----------------------------------------------------------------------
0. Concept

Kanna is a command line utility for getting size of images.  This is
inspired by Kanna from "Tamako Market", who is always seen measuring
things.  I have been writing ad-hoc scripts for doing this task, but I
figured I will get it all in one utility this time.

----------------------------------------------------------------------
1. Image formats

Depending on the list of formats supported, this kind of program can
either be very large or very small.  I have decided to make Kanna
support only the formats I used frequently, which comes down to
GIF/PNG/JPEG and the NetPBM formats.  This excludes some formats which
would have been easy to support, such as BMP and TGA.

The one format that I do use very often but decided not to support is
NEF (Nikon Electronic Format).  To start, it's a proprietary and
poorly documented format.  The next closest thing is TIFF.  After
reading much of the 120+ page spec on TIFF 6.0, I have concluded that
it's totally a writer-friendly and reader-hostile format, and not
worth the effort.  At this point I stopped looking for formats that I
might support (such as WebP) and just started coding.

----------------------------------------------------------------------
2. Image decoder

This was totally straightforward: check the first few bytes to see
what format the input is, then collect the bytes from the various
offsets to get the image size.  Some care was taken to make sure Kanna
outputs "0 0" on truncated inputs, but mostly there is nothing
excessively clever about decoding file formats, especially since we
only care about the header part and not the pixel data.

----------------------------------------------------------------------
3. Image generator

After implementing the image decoder and formatting the source into
ASCII art, I still have ~168 bytes left to spare.

Usually I would embed a BF program in the spare bytes, which I have
done (and it took only 148 bytes).  This time around, because Kanna is
relatively low-resolution, the BF program stood out too much with its
distinctively low entropy character set.

Since 168 bytes is actually quite a few bytes to spare, I threw away
the BF program and embedded a Perl program instead.  Because the main
program is an utility for determining image dimensions, it's only
fitting to have the secondary program be one that generates images of
user-specified dimensions.  This was fairly easy with Perl outputting
PBM, I was even able to fit in some random dithering stuff in it.

----------------------------------------------------------------------
4. Finally...

This was a fairly straightforward program to write, from initial idea
to the final completed ASCII art.

Nonetheless it took months to complete.  This was totally a problem
with coding time and not with coding skills.  Though I always feared
that if I stop doing something, I will lose those skills and never do
those things again.  Eventually I made time to complete this ASCII
art, and felt a lot better since.

Now I can go back to re-watching Tamako Market without feeling that
the time was wasted.