| Age | Commit message (Collapse) | Author |
|---|
|
|
|
When using our current-load/used-compiled function, we were compiling
the syntax of a module using compile, which seems to not honour
With luck, this should address bug #14 for real.
|
|
When evaluating (re)definitions in a typed module, it's necessary that
the form evaluated is wrapped with #%top-interaction, so that typed
racket's redefinition of that macro enters into play and the system
records the type information of the new value.
Many thanks to Sam Tobin-Hochstadt for the tip, and for his encouraging
words.
|
|
We used to check for a good racket version during its start-up, but
these days we already have an independent version check before that.
|
|
|
|
The new submodules and errortrace interact badly, for what i've seen.
In particular, even with the submodule[+*] loading correctly, its
namespace doesn't have all identifiers bound, and new ones seem to
appear in the bindings lists (things like a.1 or b.2, when a and b are
the actual identifiers defined inside the module).
Since moreover someone mentioned in the devel ML that errortrace is in
general terms buggy, i guess we can leave without it for the time
being.
|
|
just because we can
|
|
Submodule (re)loading is not without pecularities. In particular,
module[*+] submodules are not visited the first time one enters its
parent, but once you load them once, they're revisited every time we
load the parent afterwards--racket's native enter! exhibits the same
behaviour, so i'm guessing we'll have to live with that.
There is however a glitch in that submodules can only be reloaded then
by loading the parent, so we need to confirm that this is expected
behaviour and, if it is, automating the parent's load when the
submodule's is requested.
On the other hand, entering a module[*+] is not working in Geiser yet,
and it does in plain racket, so this one is our fault. Working on it.
|
|
... and used also internally for C-c C-k, although it doesn't yet work
as well as i wanted when it comes to load modules. The reason is
probably in geiser/enter, where we don't record modification times per
submodule but per path, which is not correct in the presence of submodules.
|
|
|
|
|
|
That is, complying to the submodule loading protocol (cf. racket's own
enter!).
|
|
And we display it (the current path, settable via ,cd) as a string.
This was, i'm sure, a secret command nobody is using!
|
|
It is now possible to ,enter racket submodules. This is only the
first part of the story, because evalations should take place in the
submodule, not in its top level module, as it happens now.
|
|
... by the obvious device of waiting for the thread building the index
to finish.
|
|
The evaluation of the help form must happen in a good enough
namespace.
|
|
|
|
For some reason that i don't fully understand, evaluating a function
in the racket/base namespace first thing after loading errortrace
breaks the help macro (!). This patches provides a workaround by
actually invoking help first thing when Geiser starts, with alibi that
it serves to preload the help index (in a separate thread).
While i was at it, i improved the message printed in the minibuffer
when no help is found.
|
|
So, the problem was that our regexp for a Racket prompt didn't take
into account that filenames could contain white spaces: "@[^ ]*> ". A
simple solution was accepting them: "@[^>]+> " won't work because '>'
is also a valid character in filenames, so we went for "@.*> ".
The drawback is that finding the beginning of the prompt (e.g. in C-a)
fails when you're writing things like:
racket@foo bar.rkt> (> 2 3)
because here comint believes that the prompt is "racket@foo bar.rkt> (> "
And that could have side-effects elsewhere. So what i've done is
simply changing the way white-space is (not) printed in the prompt,
substituting it by underscores. That way, whe can go back to the
initial regexp, comint doesn't get confused, and users can easily
infer that "@foo_bar.rkt>" is actually referring to their
"foo bar.rkt" file.
|
|
Thanks to Haiwei Zhou for catching this one!
|
|
Our module loader is receiving load requests for module names
represented as lists that are not exactly a submodule, in the sense
that the path does not represent an actual file.
This phenomenon happens for instance when specifying a reader in a
#lang tag. E.g.
#lang at-exp racket
will cause the loader to be called with module name '(main reader) and
path <cols-path>/at-exp/main.rkt, where main.rkt does not exist.
Afterwards, we see a call to load at-exp/lang/reader/rkt, with name
reader, which is the real code.
So, for now, i'm skipping all load requests with a list name,
forwarding them to racket's default loader.
|
|
Racket is returning by default their canonical "rkt" name, which
sometimes is not what's in the filesystem.
|
|
Resolved module path can now be, besides symbols, a list representing
a submodule. When deciding whether what we are loading in enter's
current loader is a module or not, we have now to take that bit into
account.
|
|
|
|
In geiser-racket.sh, there's the new option -n, which uses a new
hostname argument accepted by geiser/user's start-geiser function.
|
|
|
|
|
|
|
|
When no cache dir is set in the emacs customization, we ask Racket for
the one that it's using by default.
|
|
Brought to you by a comma-command in the REPL and the REPL startup
function.
|
|
On the racket side, we use a custom print handler to print
images (convertible? values; see file/convertible) in a special format:
#<Image: filename>
On the geiser side, we add a comint post-output hook to search for
that filename and replace it with inline images.
|
|
This one probably requires Racket 5.3, and incorporates some
parameterization to the module compilation and evaluation code in
Geiser's version of enter.rkt. I'm mostly mirroring what the latter
does, and i'm probably not completely understanding all corner cases,
so the two users of Geiser should keep an eye open for possible
breakage introduced by this patch.
|
|
It's not perfect and undocumented, but useful nonetheless.
|
|
This bugs was exposed by using rackunit, where all the output of, say,
check-eq? was lost for good (it was being sent to the stderr black
hole).
Hat tip Grant Retkke.
|
|
|
|
As Stanisław Halik reported, autodoc was receiving a mlist (instead of
a list) as its argument when invoked in an R5RS context. Turns out we
were evaluating geiser-eval's arguments in the wrong context (the
procedure being applied and the arglist belong to the same namespace).
|
|
Thanks to Caleb Reach.
We were using current-input-port, which is not the right port in
graphical environments.
|
|
And we take the chance to lightly document the existence of this
new command in the user manual.
|
|
The catch here is that one cannot use #%variable-reference inside an
R5RS module, and, as a consequence, namespace->module-path-name was
failing badly. The solution is to take note of the module name being
entered before hand, and use that name in case of error (we could
actually use that name always, but then cheaters using Racket's enter!
would see an inconsistent name (which probably they deserve)).
|
|
|
|
We avoid using elisp's read for symbols, reading uninterned ones
instead. And then, we cannot use symbols as keys in responses from
scheme: we're using strings instead.
|
|
When the symbol is imported and re-exported by a second module, we
display its definition name and original module, besides the name of
the module re-exporting it.
|
|
We have a new "manual lookup" command, and Racket now displays a doc
browser buffer for help with a button activating it. In the process,
we've cleaned-up a little mess in geiser-eval.el and geiser-doc.el,
and refactored the affected Racket modules.
Next in line is providing manual lookup for Guile.
|
|
|
|
Refreshing the cache on form evaluation (had been deactivated by error
since the introduction of meta-commands). The current behaviour is not
yet completely correct: if one evaluates a form in a modified buffer,
geiser won't notice the new definition's signature, not even after the
buffer is saved if one has gone around asking for autodoc before that.
An improvement would be to make the cache sensitive to file write
times, as suggested back in the day by Eli & Neil.
|
|
We now display procedure signatures in module help, and keep a cache
in Guile, using procedure properties.
|
|
Spinning up from correct fontification of [else in this brave Racket
world.
I'm keeping the list of extra keywords lean and mean, but making it
customizable in both Racket and Guile.
|
|
Inferior schemes weren't really a good idea, were they? With remote
connections one can launch an external scheme to debug Geiser anyway.
And everything is (ahem, will be) simpler when we add new
implementations.
|
|
Separate connections for the REPL and Geiser commands was kind of
neat, but it had the problem of synchronising the current namespace
for both connections. A quick fix would have been to ask the scheme
for the current namespace for every Geiser command in the REPL, but
that, besides clunky, would add potentially prohibitive overhead for
(real) remote connections.
As it happens, using a single connection turned out to be not that
difficult and relatively clean code-wise. We could even turn back to
not use inferior schemes, and the net result of this refactoring would
be the replacement of comint-redirect (which wasn't able to match the
whole EOT token if it didn't arrive all at once) by transaction queues
(which also makes geiser-connection's implementation cleaner).
But using an inferior scheme has a dog-food value, and allows external
processes to connect to the scheme being used by Geiser without
further ado, which could be useful for debugging (although this is a
lame excuse: nothing prevents you from starting a REPL server from
emacs if you want). We'll see.
|
|
|
