closes PR 11220
Altho, this does not fix the case where a function is being passed thru
another contracted function. Eg, when you give the identity function
this contract: (-> (-> integer? integer?) (-> integer? integer?))
if you pass some function with a name in there, it won't come
back with a name anymore (indeed, it won't even have the name
anymore in the body of the function).
For that, the fix would have to be put into each of the function
contract combinators.
[It will not bother me if we revert this commit. I liked SK's idea and found it easy to implement. I wonder if others will be worried that it is easy to unintentionally leave off the second argument to check-error. I also wonder if it is problematic to add new string constants, like I've done.]
Here is an example:
(check-error (/ 1 0) "/: division by zero")
(check-error (/ 1 0) "divide by zero")
(check-error (/ 1 0))
(check-error 1)
Here is the output:
Ran 4 tests.
2 of the 4 tests failed.
No signature violations.
Check failures:
check-error encountered the following error instead of the expected divide by zero
:: /: division by zero
in ex.rkt, line 2, column 0
check-error expected an error, but instead received the value 1.
in ex.rkt, line 4, column 0
Create a mzlib/contract compatible version of the old box/c and use that
for mzlib/contract. Change the docs so that the docs for mzlib/contract
contain the right information.
Fix the typed-scheme implementation to only force flat box (or hash)
contracts when it already is required to be flat. Otherwise, allow
non-flat contracts for the element contract (or domain/range contracts).
* Domain contracts must be either flat or chaperone contracts.
* When the domain is a chaperone contract, hash/c must check
that the hash is equal?-based.
* Flat contracts are chaperone contracts, and chaperone contracts
are (proxy) contracts.
* Check in chaperone contracts that a chaperone (or chaperone-friendly
value) is indeed returned.
- when optional arguments didn't also have rest args,
it generated the wrong wrapper function
- the object-contract interaction was broken;
it was not adding the this argument in
Also removed the hack added earlier this eve.
closes PR 11180
Now use for/fold to thread the index through the iteration form, so
that all variants can take advantage of the optional #:length
argument. Previously, only the for/vector and for/flvector used the
The behavior when #:length does not match the number of iterations has
changed: iteration stops when either the vector is full, or the
requested number of iterations has been achieved, whichever comes
first. If #:length is larger than the number of iterations performed,
then the remaining slots in the vector are filled with the default
argument of (make-vector ...), which is currently 0.
- Now the faster forms take a #:length keyword to designate the length
of the vector to pre-allocate.
- The for/[fl]vector forms take multiple body expressions and set the
vector component to the value of the last one.
- When given a #:length argument, the for/vector and for/flvector
forms check that the iteration is not exceeding the given length,
raising exn:fail if it does.
- Test cases for the multiple body expressions and the exception for
excessive iterations have been added.
- Doc modifications to bring the docs in line with the new forms.
- Doc modifications to note that the #:length versions of the form
*may* all the computation to be performed more efficiently, and
stating that it "is an error" if the given length-expr does not
produce a valid length for a vector that matches the number of
iterations for the loop.
- Note that no test is made for a number of loop iterations that is
smaller than the given vector length. Also, the for*/[fl]vector
forms do not optimize when given a #:length argument. These are
areas for future improvement.
Changed backtracking algorithm, runtime representations
- syntax classes, ~describe no longer implicitly commit
- ~describe no longer delimits effect of cut
Added keyword & optional args for stxclasses
Added ~do and #:do, ~post, ~commit and #:commit,
~delimit-cut and #:no-delimit-cut
Added syntax/parse/debug, syntax/parse/experimental/*
- expr/c for contracting macro sub-expressions
moved from syntax/parse to syntax/parse/experimental/contract
- syntax class reflection (~reflect, ~splicing-reflect)
- eh-alternative-sets (~eh-var)
- provide-syntax-class/contract
(only for params, not attrs so far)
Changed ~fail to not include POST progress (#:fail still does)
old (~fail _) is now (~post (~fail _))
Made msg argument of ~fail optional
Removed generic "repetition constraint violated" msg
Removed atom-in-list stxclass
Removed unnecessary datum->syntax on cdr of pair pattern
massive improvements to long-list microbenchmarks
Optimization: integrable syntax classes (id, expr, keyword)
need better measurements
Optimization: ad hoc elimination of head/tail choice point
for (EH ... . ()) patterns
Added unstable/wrapc (proc version of expr/c)
