Support `@constraints m [i in 1:5] begin ... end`

[cstjean]

Is there any reason not to support that syntax? I have a big block that looks like

@constraints m begin
    [i in 1:5], ...
    [i in 1:5], ...
    [i in 1:5], ...
    [i in 1:5], ...
    [i in 1:5], ...
end

It'd be nicer if [i in 1:5] appeared only once.

@constraints m [i in 1:5] begin
    ...
    ...
    ...
    ...
    ...
    ...
end

I'm assuming that it was rejected for some reason (and thus please close this issue), but if not, I believe I could make a PR for it.

[odow]

Off the top of my head, the main reasons not to allow this:

1. it makes the macros more complicated, and they are already very complicated
2. What is the interaction with registered constraints? Does this apply only to anonymous constraints?
3. Can a row have other indices?
4. What is the return value? Is it a vector of constraint tuples, or a tuple of constraint vectors?

Here's an example to think about:

model = Model()
@variable(model, x[1:5, 1:2])
ret = @constraints(model, [i in 1:5], begin
    c, sum(x[i, :]) >= 0
    [j in 1:2], 2 * x[i, j] <= 1
end)
typeof(ret) # ???

The recommended approach would be:

for i in 1:5
    @constraints(model, begin
        ...
        ...
    end)
end

[odow]

Also: how does this work with @variable and @expression?

[cstjean]

it makes the macros more complicated, and they are already very complicated

I can relate to that! I was under the impression that

for i in 1:5
    @constraints(model, begin
        ...
        ...
    end)
end

was somehow slower, but apparently that's not the case. Alright, I agree that the trade-off doesn't look favorable.

[odow]

was somehow slower, but apparently that's not the case

Nope. There are cases where it may even be slightly faster because you're not creating or storing the resulting containers.