Altered TR's random arithmetic testing to generate single-flonums and very small flonums; fails now because of erroneous types
Fixes to sgn, sinh, cosh, and tanh:
* preserve single-flonum-ness
* correct zero sign (-0.0) for negative return values that are smaller than epsilon
* correct behavior with NaN and infinite inputs
in the bottom window when it really is a different file
(ie, using a same-file predicate that involves opening
the files and comparing the ports)
closes PR 12635 (hopefully -- if not, then we can reopen)
Caveats:
- keyword function definitions do not define static
bindings, thus limiting optimization opportunities
- can't use `define:`, `lambda:`, etc with keywords
- error messages sometimes expose the implementation
- the optimizer skips most of the generated code for
keyword functions definitions (user-level code is
optimized)
This reverts commit fc8cb6e37d.
Jay changed the set of installed font packages on DrDr to match the
ones on my linux VM which brought the test results back in sync again,
so we don't need these revised bitmaps anymore.
Thanks, Jay!
Also, fix FFI procedures to preserve names: change `ptr-ref' with
`_fpointer' on an `ffi-obj' value to return the `ffi-obj'
value, so that the name in the `ffi-obj' value can be used
by `_cprocedure'.
Closes PR 12645
Previously, some hacks were used to obtain the internal
identifiers that implemented keyword functions directly, and
give them types at startup. Now, the primary "function"
(eg, `sort`) is given a type, and when used, the residual
syntax properties are used to find `sort` from the real
functions, and then the type of the real function is computed
from the type of `sort`.
Some creativity was required in the types of functions which
take optional arguments that when present, alter the return type,
such as `regexp-match*` and `file->list`.
Specifically:
- make predicate symbols and variables be original (and,
in the former case, have srclocs)
- remove the originality in the quoted syntax constants
that the expansion introduces
I think there is probably something better that should be done
with those introduced syntax constants, possibly relacing
them with #f, or possibly replacing them with source location
lists, but I'm not sure how they are used, so I've just left
them in there.
(Also, very minor Rackety in check syntax)
the error raising code, except in one place where I think the
error raising code was correct to have the plural form, so I changed
it back instead of changing the Redex test suite to match
like textual-read-eval-print-loop
There is still a difference, however, because drracket's REPL has a
notion of multiple expressions that are submitted simultaneously that
textual-read-eval-print-loop doesn't. For example, if you type this at
the prompt:
(car) (+ 1 2)
then textual-read-eval-print-loop will print out the error and then 3,
but drracket will print only the error (ditto if (car) were replaced
by a continuation abort).
This difference is, IMO, a good thing, since it lets you use a single
interaction to do multiple things, but stops as soon as there is an
error. (It is also how drracket has behaved for a long time.)
closes PR 12790
