This will allow us to merge fixes with greater confidence going forward.

incl. minimal documentation and a pair of tests

Also replace use of `count` with `counting` in other tests.

Steve Losh authored Nov 29, 2019 and Robert Goldman committed May 18, 2021

Control flow clauses that return from an iterate form without executing
the epilogue code (e.g. `never` and `leave`) currently do so by
expanding into `(return-from ,*block-name*)`.  For iterate forms without
a user-provided name, the block will be named `nil`.  This causes
problems when used with other Common Lisp looping constructs such as
`dolist`:

    (iterate
      (for i :from 0 :below 10)
      (dolist (x '(5 10 15))
        (never (= i x))))
    ; => T, but should be NIL

This patch adds a separate `*loop-name*` variable that will be bound to
a gensym for `NIL`-named blocks and makes the various epilogue-skipping
code refer to that instead of `*block-name*`.  Iterate forms *without*
an explicit user-provided name will still have a separate `(block nil
...)` wrapping so users can still `(return ...)` as before.

Before (simplified):

    (iterate (leave))
    ; ==> (block nil (return-from nil))

    (iterate foo (leave))
    ; ==> (block foo (return-from nil))

After:

    (iterate (leave))
    ; ==> (block nil (block #:ITERATE123 (return-from #:ITERATE123)))

    (iterate foo (leave))
    ; ==> (block foo (return-from foo))

A spurious warning is generated by ITERATE when adding a type
declaration for driver variables.  This is caught by test `type.8`.

There was a cut-and-paste typo in this test.

Robert Goldman authored Apr 17, 2020 and Attila Lendvai committed May 11, 2020

On unmodified version of iterate on CCL this test will fail.

Ignore-this: 47c28331db5d7aa76625078f3ff5f9bb

darcs-hash:8e3164ce4db2468289cd426aa8a791d809df62ea

Ignore-this: f9aae702d7b6cad7fed86ab8a195c45d

darcs-hash:7a69bd70258432079c1daee2c119fe3e7ee87048

Ignore-this: 7d072a9ffe694700e1dc52ef3c653f94

darcs-hash:79473b46ee0df4304125a8727b0691cfb05bbdd7

Ignore-this: f3e9d3983567eb04ce2429b310bee837

I'm not sure about what is their contract, but it was surprising to me
that the FINALLY block is not executed.

Arguably if you're expected to avoid sideffects in a FINALLY block,
then you cannot/shouldn't change the return value, so the entire
FINALLY block can be skipped then.

Either way, this behavior should be documented with the tests, so
if the current behavior is expected, then change the tests
accordingly (as opposed to deleting them).

darcs-hash:c0b172b51bc40943946c86d8ab76297807d26594

Ignore-this: 6e9f419e118724c660d519c3fa9f8dd2

darcs-hash:2c0bdf6f95d87e1369aaa184220cf96fecbf0f39

Ignore-this: d55e7c22deeaf89b90d03d7ef01179d6

darcs-hash:0384e9304712c6ce86921411449adaebcab39fc2

Ignore-this: faa1bd48d0450e76652552bb47bcff02

darcs-hash:3014735c0464ffdf47d05a0f8dd5ad6fc1f0db9b

darcs-hash:bc746991bd3bbb39b86b4249d2bb71bf6d8f9a3e

More precisely, FOR ON accepts dotted lists, FOR IN errors out.
As a result, iterate::*list-end-test* was eliminated.
Behaviour is now constant and does not depend on some special variable.
Note: Documentation not yet updated, pending move to Texinfo.

darcs-hash:5d57aef5caed4ed28274d852bd8626f19c3d0c94

darcs-hash:0a41a0f97d7b6d96454f41c0f79e560bc5c74cd0

darcs-hash:07fba405cd24f8b6628a262c397d500114dbe842

darcs-hash:68770ce3bdc7525dc7f9fcb9100b3895c8c488c6

darcs-hash:f9cc3e39001d6f290d63a72ddfb32a9b47382794

darcs-hash:60aa3d2a88d512f3d26c6a6623c25721a02bbf4e