Detect cycles in the toplevel's printer.

[lpw25]

The patch attached to issue 6228.

[gasche]

the short branch should always come first in an if .. then .. else

[gasche]

One potential issue with the proposed approach is that all cycles are printed in the same way. If I understand correctly, it is not possible to distinguish the pretty-printing of let rec x = 1::2::x and 1 :: let rec twos = 2 :: twos, both would return 1 :: 2 :: <cycle>.

The old patch by Alain on the bugtracker had the advantage of keeping track of cycle identities, and properly adding a bound name at the first occurence of a cycle. It would print something like let rec v1 = 1::2::v1 in one case and let rec v1 = 2::v1 in 1::v1 in the other.

The downside with this is that it enters the nightmare of having several possible syntaxes which makes decision-making impossible. It also looks like a more invasive change to the printer implementation (with the current style of nest combinators I would expect that inserting the fixpoint at the first point of occurence, rather than at the toplevel, to be easier; that would also be a fine semantics).

[lpw25]

One potential issue with the proposed approach is that all cycles are printed in the same way. If I understand correctly, it is not possible to distinguish the pretty-printing of let rec x = 1::2::x and 1 :: let rec twos = 2 :: twos, both would return 1 :: 2 :: .

I don't think this really matters, in the top-level you can always perform some more operations to get a better idea of what the value is. The important thing is just not to print hundreds of lines of output for simple values.

[lpw25]

I've reordered the if clauses, made a few cosmetic changes, and fixed a bug in the printing of abstract types with manifests, Tpoly types and Tsubst types.

[vouillon]

Note that using the same hash for many values will result in a quadratic complexity for hash-table look-up.

[lpw25]

@vouillon I assume you are referring to the handling of exceptions raised during Hashtbl.hash. This should only occur if someone installs a custom hash function which raises an exception, which should be extremely rare, so I'm not really worried about performance in that case.

[mshinwell]

Merged to trunk (rev 14766)