Generate table of overloads in docs

[adrhill]

Closes #115.

Generates more granular tables than issue #115.
Inside of the tables, checkmarks link to the respective source code.

Caveat: this doesn't include manual links to open issues.

Preview

Table of overloads.pdf

[gdalle]

bro that rocks

[gdalle]

Why is the table not in alphabetical order?
Also, FastDifferentiation is the only backend for which everything is overloaded, so it should be all green?

[adrhill]

Why is the table not in alphabetical order?

If you remove the value_and_ functions, it is.
I can swap operator! and operator if you prefer.

Also, FastDifferentiation is the only backend for which everything is overloaded, so it should be all green?

Good catch, that's a bug.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 96.39%. Comparing base (32849b8) to head (b5b2538).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #207   +/-   ##
=======================================
  Coverage   96.39%   96.39%           
=======================================
  Files          65       65           
  Lines        3554     3554           
=======================================
  Hits         3426     3426           
  Misses        128      128           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[gdalle]

I would put more stuff in a single line, like all four variants of gradient for example

[adrhill]

Does this look correct for FastDifferentiation?

[adrhill]

I would put more stuff in a single line, like all four variants of gradient for example

What would you call the columns?

[gdalle]

No, it also has HVP. The source of truth is the GitHub issue

[gdalle]

What would you call the columns?

For each backend, one table for 1-arg functions f and one table for 2-arg functions f!

In each row, a single operator like :gradient, and four columns with headers:

• op
• op!
• val_and_op
• val_and_op!

Alternately, the two tables could be brought into one

[adrhill]

The problem with hvp! and hvp was the missing vector argument.
There are too many edge cases, so now I manually specify operator function signatures for both f(x) and f!(y, x).

Updated preview

Table of overloads.pdf

[gdalle]

The problem with hvp! and hvp was the missing vector argument.

Beware, pushforward and pullback also take one

[adrhill]

Round 3: Table of overloads.pdf

[gdalle]

It looks better at first glance, I'll check in detail tomorrow! We probably don't need the 2-arg table for backends that don't support these functions, if that's not too much trouble to remove

[adrhill]

Good call, I've now filtered them. It makes the tables a lot more readable.
I've also added a couple of sentences to explain to people what they are looking at.

Latest build: Table of overloads.pdf