Port to Gerbil Scheme

[vyzo]

This adds support for Gerbil in the Scheme implementation.

[kanaka]

@vyzo can you add the installation of gerbil to the Dockerfile and add that mode to .travis.yml to activate Travis testing? Once you do that I'll rebuild and push the docker image and restart the Travis test.

[vyzo]

sure!

[vyzo]

It seems the test failed because gxc not found. Have you updated the docker container?
Perhaps I should install in /bin instead of /gerbil/bin and adding to path in the env.

[vyzo]

I updated the Dockerfile to install gerbil in /opt/gerbil.

[wasamasa]

@vyzo I planned to wait with this one until Gerbil scheme has been properly packaged so that no installation to /opt or alike is needed and binaries would end up in /usr/bin like with everything else.

• The gerbil-build.sh thing looks mostly useless. Why not make it part of the Makefile? Instead of building everything at the first step, you build/rebuild stuff as needed. The existing definitions show how it's done for other Scheme implementations. Note that $(RM) -rf isn't needed, it's covered by $(RMR).
• What's the deal with (export main) and the later (interactive-main)? Is there a way to disable this special interpretation of the main procedure? Do you really have to rename the main function because it has a different function signature from the exported main procedure?

[vyzo]

I can change the Dockerfile to install on /usr/bin, that's not such a difficult thing.

• The gerbil-build.sh script is used to build because the gerbil compiler output doesn't match the 1-to-1 model of Makefiles; there are several artifacts produced by the compiler for each module and it was much simpler for me to just have a script than trying to fight with the Makefile. Sorry, my Makefile fu is not strong enough.

• Gerbil requires a main function as entry point for executables, which receives the command line arguments. There is (currently) no way to disable this behaviour, which is desirable for several reasons (and you can also argue that it's undesirable for some other reasons).

[kanaka]

@vyzo I pushed out an updated build of the scheme docker image and restarted the gerbil task. There are some errors during the build phase and also seem to be some ASCII control characters being printed in the output that are breaking the tests: https://travis-ci.org/kanaka/mal/jobs/357610551

[vyzo]

seems my build script is not expanding the lib/{util,types,env,reader,printer,core}.sld construct -- probably because the underlying shell is not bash.
Let me try to fix it.

[vyzo]

not sure what's up with the control characters though.

[vyzo]

I also recently introduced a bug in Gerbil r7rs, which is now fixed:
mighty-gerbils/gerbil@aa051ea

We will need to build a new docker container for that fix, as it breaks mal compilation.
Sorry about that.

[kanaka]

Hi @vyzo, are you still interested in pursuing this? I built and pushed a new scheme docker image that updates the gerbil repo as of today it's state a couple hours ago. The first thing I ran into is that the build rule dependencies aren't complete. Here is a patch to my tree that fixes that (and also moves the gerbil-build.sh logic into the Makefile):

diff --git a/Makefile b/Makefile
index d88cd89..0ba8e49 100644
--- a/Makefile
+++ b/Makefile
@@ -169,6 +169,7 @@ scheme_STEP_TO_PROG_chicken     = scheme/$($(1))
 scheme_STEP_TO_PROG_sagittarius = scheme/$($(1)).scm
 scheme_STEP_TO_PROG_cyclone     = scheme/$($(1))
 scheme_STEP_TO_PROG_foment      = scheme/$($(1)).scm
+scheme_STEP_TO_PROG_gerbil      = scheme/gxc.out/bin/$($(1))
 
 # Map of step (e.g. "step8") to executable file for that step
 ada_STEP_TO_PROG =     ada/$($(1))
diff --git a/scheme/Makefile b/scheme/Makefile
index ccb719e..94b86f5 100644
--- a/scheme/Makefile
+++ b/scheme/Makefile
@@ -9,11 +9,11 @@ CLASSSTEPS = out/step0_repl.class out/step1_read_print.class \
              out/step3_env.class out/step4_if_fn_do.class out/step5_tco.class \
              out/step6_file.class out/step7_quote.class out/step8_macros.class \
              out/step9_try.class out/stepA_mal.class
-GXCSTEPS = gxc.out
+GXCSTEPS = $(foreach s,$(BINS),gxc.out/bin/$(s))
 STEPS = $(if $(filter kawa,$(scheme_MODE)),$(CLASSSTEPS),\
         $(if $(filter chicken,$(scheme_MODE)),$(BINS),\
         $(if $(filter cyclone,$(scheme_MODE)),$(BINS),\
-               $(if $(filter gerbil,$(scheme_MODE)),$(GXCSTEPS)))))
+        $(if $(filter gerbil,$(scheme_MODE)),$(GXCSTEPS)))))
 
 KAWA_STEP1_DEPS = out/lib/util.class out/lib/reader.class \
                   out/lib/printer.class out/lib/types.class
@@ -71,8 +71,12 @@ eggs/r7rs.so:
        chicken-install -init eggs
        CHICKEN_REPOSITORY=$(CURDIR)/eggs chicken-install r7rs
 
-gxc.out:
-       ./gerbil-build.sh
+gxc.out/lib/core__rt.ol:
+       mkdir -p gxc.out/bin
+       GERBIL_LOADPATH=gxc.out gxc -d gxc.out -O lib/util.sld lib/types.sld lib/env.sld lib/reader.sld lib/printer.sld lib/core.sld
+
+gxc.out/bin/%: gxc.out/lib/core__rt.ol
+       GERBIL_LOADPATH=gxc.out gxc -d gxc.out -O -exe -o $@ $*.scm
 
 lib/%.scm: lib/%.sld
        $(SYMLINK) $< $@

Once that is applied, the steps will build and the tests can run but then I see the original failure I was seeing where there is escape codes in the output. Here is a snippit of the test failure (test^step0):

Testing basic string
TEST: abcABC123 -> ['',abcABC123] -> FAIL (line 3):
    Expected : 'abcABC123\r\nabcABC123'
    Got      : '\x1b[1mabcABC123\x1b[m\r\nabcABC123'
Testing string containing spaces
TEST: hello mal world -> ['',hello mal world] -> FAIL (line 7):
    Expected : 'hello mal world\r\nhello mal world'
    Got      : '\x1b[1mhello mal world\x1b[m\r\nhello mal world'

If you're still interested can you fix the failures? Also, since @wasamasa created the base scheme implementation he'll either need to sign off on the way you are doing the main invocation, or you'll need to bring that more into line with the way all the other scheme implementations currently work.

If you're not interested in pushing this forward, please let me know and I'll close the issue. Thanks!

[vyzo]

I can look into fixing the test failures.

[vyzo]

Seems like output is getting garbled with control characters.

[kanaka]

I suspect it's whatever library you are using to do line editing (e.g. readline, libedit, linenoise) is trying to do some sort of input highlighting or cursor movement using escape codes.

[kanaka]

@vyzo Any interest in revisiting this and getting it merged?

If so, you'll want to do a few things since so much time has passed:

• rebase onto the HEAD sources
• move the .travis.yml definition to IMPLS.yml
• Make sure that the Github CI tests all pass for your implementation. The process is now more automated and doesn't require my manual intervention to build docker images (part of the reason to rebase).

Thanks!

[vyzo]

Sure! It's been a while, and Gerbil has evolved a lot since, but it could be a fun exercise.

Time is not ample unfortunately, but i'll put it in my list of things to hack.