There used to be a comment here with a bit of explanation of what parameter is what. I think that's still a good idea here (but possibly a completely different comment because behaviour changed).

End sentences with a period. The comment could also be put on two lines, I'm guessing.

This block is indented too much. If pchan == NULL or conlist == NULL you can just return. This saves two indent levels.

This should be a switch statement, and not a sequence of if/elseif.
Every case should be its own properly-named function, with clearly named parameters.

The next code block (i.e. empty-line-delimited bit of code) doesn't mirror anything, so the comment is a bit weird here.

What does it mean to "represent X"? This code seems to find the matrix row with the largest first value, but it's not clear why this is even relevant.

Where is this mapped? What defines those values? Where do they come from? So many magic numbers.

Why not write min_vec[0] = -min_vec[0]? Or write min_vec[0] *= -1? Personally I prefer the latter, as it's clear at first glance that it's only taking the negative (so no swapping min/max or changing the index).

Just declare int i as part of the for-loop (so for (int i=0; ...). Outside of a for-loop it's a rather horribly-named variable.

Just *= -1 those

act doesn't change within the for-loop, so you can do the DEG update outside the loop.

It's not clamping, it's modulo'ing. And yes, that's a good thing to do. In a separate function, in BLI_whatevermath.

Once working on it anyway, you may want to add two normalisation functions, one producing a result on the interval [-180°, 180°), and one on [0°, 360°).

Add something like "This code assumes that bRotLimitConstraint and bLocLimitConstraint have the same fields in the same memory locations."

Possibly it would be prudent to add a similar warning to the types themselves, pointing back at this code.

Don't use mat or imat, but give them a descriptive name, something that is a bit closer to "constraint space conversion matrix".

Unclear why the 'min vector' and 'max vector' are relevant concepts at all.

Bit of a weird choice in empty lines here, possibly better grouping possible?

tar → target

own, imat → something more descriptive

Why not just = -old_max?

Don't name things temp, but describe what it is. The fact that it's here in an if in a loop in an if already shows that's temporary.

Declaring where it's needed and then naming descriptively ('copy of scale' is likely better than 'temp vector') is better.

switch!

You could even go further and de-uglify it locally a bit by #define AXIS_X 0 and the like (or use a const char for that). That way at least we have some understandable word there, instead of just a 0.

Same as above, just do if (ob->pose == NULL) { return; }

Code style: use underscore instead of camel case

Same as before, probably can use continue here. Patch wasn't submitted with Arcanist, so I can't see outside it (no context available), so maybe this is not the case.

Shouldn't this use head[axis] instead?