Failure running the agent and client on ABCL

[Erik Huelsmann]

Hi,

I'm trying to run the agent on ABCL, but the command it sends to the ABCL test client doesn't do anything. This is the command it sends to the client ABCL:

java -XX:MaxPermSize=256m -jar ~/src/abcl-trunk/dist/abcl.jar --noinit --nosystem --batch --eval '(LOAD #P"/home/ehuelsmann/src/cl-test-grid/agent/proc-common.lisp")' --eval '(SET-RESPONSE #P"/home/ehuelsmann/src/cl-test-grid/agent/../work-dir/agent/response7647489212893815998.lisp" "abcs *() { /d e11 ")'

Any idea where I should start looking?

Bye,

Erik.

[Anton Vodonosov]

Hello Erik,


11.08.2012, 13:53, "Erik Huelsmann" <ehu...@gmail.com>:
> Hi,I'm trying to run the agent on ABCL, but the command it sends to the ABCL test client doesn't do anything. This is the command it sends to the client ABCL:

> java -XX:MaxPermSize=256m -jar ~/src/abcl-trunk/dist/abcl.jar --noinit --nosystem --batch --eval '(LOAD #P"/home/ehuelsmann/src/cl-test-grid/agent/proc-common.lisp")' --eval '(SET-RESPONSE #P"/home/ehuelsmann/src/cl-test-grid/agent/../work-dir/agent/response7647489212893815998.lisp" "abcs *() { /d e11 ")'
>
> Any idea where I should start looking?

This command is executed in the check-config phase to ensure agent can communicate with
subordinate lisp implementations. It instructs child ABCL process to returns result string "abcs *() { /d e11" and agents
tests if he can receive this string.

Do you have any errors? There is a log file in cl-test-grid/work-dir/agent/logs, you can check there.

Actually, I advice to utilize CCL for running the main process - agent. It will save your time in
debugging (so that you will not be districted from more important things like ABCL 1.1 release).
Test-grid-agent depends on several libraries, and may have portability problems.
In particular, there is an external-program library, and test-grid agent uses
external-program:start - an asynchronous execution of external process. According
to external-program documentation external-program:start is not implemented
for ABCL.

[Erik Huelsmann]

> Hi,I'm trying to run the agent on ABCL, but the command it sends to the ABCL test client doesn't do anything. This is the command it sends to the client ABCL:

> java -XX:MaxPermSize=256m -jar ~/src/abcl-trunk/dist/abcl.jar --noinit --nosystem --batch --eval '(LOAD #P"/home/ehuelsmann/src/cl-test-grid/agent/proc-common.lisp")' --eval '(SET-RESPONSE #P"/home/ehuelsmann/src/cl-test-grid/agent/../work-dir/agent/response7647489212893815998.lisp" "abcs *() { /d e11 ")'
>
> Any idea where I should start looking?

 

[snip]

 

Actually, I advice to utilize CCL for running the main process - agent. It will save your time in
debugging (so that you will not be districted from more important things like ABCL 1.1 release).

Well, the reason I want to run the test grid program is to see how much better ABCL 1.1 is doing than ABCL 1.0. I'm expecting it to do a fair bit better, but I want data to verify.

 

Test-grid-agent depends on several libraries, and may have portability problems.
In particular, there is an external-program library, and test-grid agent uses
external-program:start - an asynchronous execution of external process. According
to external-program documentation external-program:start is not implemented
for ABCL.

On another thread, Stas Boukarev provided a comment that Drakma can be loaded without HTTPS support by adding :drakma-no-ssl to the feature list - that should alleviate the problem of abcl not being able to load compiled cl+ssl fasls. (I've registered ticket #235 for that.)

Also, I think implementing what's required for EXTERNAL-PROGRAM should be doable; I'll have a look to see if I can contact Greg Pfeil (the author of EXTERNAL-PROGRAM, right?) to agree upon a good API to support it.

Bye,

Erik.

[Anton Vodonosov]

Hello Erik.

I've build ABCL svn HEAD (revision 14092) and published test results for it.

On this table, columns of the old and the new ABCL versions are side by side:
http://common-lisp.net/project/cl-test-grid/pivot_lib_ql-lisp.html

At first sight not regressions comparing to the previous ABCL, and several
improvements.

Best regards,
- Anton

[Erik Huelsmann]

Hi Anton,

Thanks for the report on ABCL-1.1. I've reviewed a number of failing libraries and found that some have incorrect expectations of standard functions. E.g. Anaphora thinks the truename of #p"/tmp/" is #p"/tmp/", but since you run on Windows, the canonical name is "C:/tmp/" which mismatches the test expectation.

I'll be reviewing more of them and contacting library authors as I find cases which should be fixed by them. Hopefully there are a number of failing tests we can fix as ABCL developers as well.

Again, thanks for the test results!

Bye,

Erik.