Ok, I think I almost understand the second example (but why does it use test_amny rather than test_equal?), but not the first one. What are the criteria for saying if the first test succeeded or failed?

/ Niels Möller (ny flexiblare bröstkorg)

Previous text:

2003-04-15 11:42: Subject: Starting new pike processes from the testsuite

---

Excerpts from the main testsuite.in:

test_do([[ // This is a memleak test rm("testsuite_test.pike"); Stdio.write_file("testsuite_test.pike", "constant Q=10000000000000000;\n" "int main() {}\n"); Process.system(RUNPIKE +" testsuite_test.pike"); rm("testsuite_test.pike"); ]])

test_any([[object p=Process.create_process(Process.split_quoted_string(RUNPIKE)+({"-e","exit(2)"})); sleep(10); return p->wait()]],2)

/ Henrik Grubbström (Lysator)

Would something like this work?

test_do([[ object p = Process.spawn(RUNPIKE, "SRCDIR/https.pike"); sleep(5); /* Wait a little for the server to startup */ HTTP.Query q = HTTP.get_url("https://localhost:25678"); if (q->status != 200 || search("html", q->data()) < 0) error("Failed\n"); p->kill(); ]])

The sleep is ugly, perhaps one could hack https.pike to send some signal to its parent once it has bound the port.

/ Niels Möller (ny flexiblare bröstkorg)

Previous text:

2003-04-15 13:08: Subject: Starting new pike processes from the testsuite

---

test_do tests fail if they fail to compile or throw an error. I don't think the intention here is for the test to actually fail if there is a memleak, but just to get lots of output on stderr.

/ Marcus Comstedt (ACROSS) (Hail Ilpalazzo!)

And make clean fails too. Adding a target

clean: true

to refdoc/Makefile helps.

/ Niels Möller (ny flexiblare bröstkorg)

Previous text:

2003-04-15 14:53: Subject: Starting new pike processes from the testsuite

---

Unfortunately, now make verify fails on one earlier test:

...

array_sscanf("\1000\1001\2000","%[\1000-\1111]%s");

make[2]: *** [run_hilfe] Segmentation fault make[2]: Leaving directory `/home/nisse/hack/pike/build/linux-2.4.18nisse-i686' make[1]: *** [compile] Error 2 make[1]: Leaving directory `/home/nisse/hack/pike' make: *** [run_hilfe] Error 2

Does anybody else get that? I'll try a make clean and a recompile first.

/ Niels Möller (ny flexiblare bröstkorg)

z is a C99 format character that is used from printing "size_t" typed values. PRINTSIZET should only get set to "z" if your C compiler handles it.

/ Marcus Comstedt (ACROSS) (Hail Ilpalazzo!)

Previous text:

2003-04-15 15:03: Subject: Starting new pike processes from the testsuite

---

And another oddity: I get

/* set this to the modifier type string to print size_t, like "" or "l" */ #define PRINTSIZET "z"

/* set this to the modifier type string to print ptrdiff_t, like "" or "l" */ #define PRINTPTRDIFFT "t"

in machine.h, but then gcc complains

/home/nisse/hack/pike/src/gc.c:2978: warning: unknown conversion type character `z' in format /home/nisse/hack/pike/src/gc.c:2978: warning: unknown conversion type character `z' in format

What's z supposed to mean?

/ Niels Möller (ny flexiblare bröstkorg)

The configure test is correct; the problem is that gcc's argument checker doesn't know about it.

/ Henrik Grubbström (Lysator)

Previous text:

2003-04-15 15:10: Subject: Starting new pike processes from the testsuite

---

Then it seems the configure test is broken. I'm currently using gcc-2.95.4.

/ Niels Möller (ny flexiblare bröstkorg)

s/C compiler/libc/

/ Henrik Grubbström (Lysator)

Previous text:

2003-04-15 15:06: Subject: Starting new pike processes from the testsuite

---

z is a C99 format character that is used from printing "size_t" typed values. PRINTSIZET should only get set to "z" if your C compiler handles it.

/ Marcus Comstedt (ACROSS) (Hail Ilpalazzo!)

I'm a little confused about how SRCDIR works in the testsuite files, and also about which of the Process.*-functions I should use. I'm trying with

test_do([[ object p = Process.Process(RUNPIKE + " "SRCDIR/https.pike""); sleep(5); /* Wait a little for the server to startup */ [...]

but that fails with

Pike: Couldn't find script to execute ("/home/nisse/hack/pike/build/linux-2.4.18nisse-i686/SRCDIR/https.pike")

/ Niels Möller (ny flexiblare bröstkorg)

Previous text:

2003-04-15 13:51: Subject: Starting new pike processes from the testsuite

---

Would something like this work?

test_do([[ object p = Process.spawn(RUNPIKE, "SRCDIR/https.pike"); sleep(5); /* Wait a little for the server to startup */ HTTP.Query q = HTTP.get_url("https://localhost:25678"); if (q->status != 200 || search("html", q->data()) < 0) error("Failed\n"); p->kill(); ]])

The sleep is ugly, perhaps one could hack https.pike to send some signal to its parent once it has bound the port.

/ Niels Möller (ny flexiblare bröstkorg)

Ah. Is the same true for RUNPIKE?

I think it would be preferable for the testsuite driver to define some functions pike_spawn, pipe_popen, that does the right thing.

/ Niels Möller (ny flexiblare bröstkorg)

Previous text:

2003-04-15 15:50: Subject: Starting new pike processes from the testsuite

---

SRCDIR is an m4 marco, so it won't get expanded inside [[]].

/ Marcus Comstedt (ACROSS) (Hail Ilpalazzo!)

Now I've changed the code to

test_do([[ object p = Process.Process(RUNPIKE + " "]]SRCDIR[[/https.pike"");

but it still doesn't work. I think I'm done for today, I'll just dnl out the test and check it in, in case somebody else would like to try it.

/ Niels Möller (ny flexiblare bröstkorg)

Previous text:

2003-04-15 15:50: Subject: Starting new pike processes from the testsuite

---

SRCDIR is an m4 marco, so it won't get expanded inside [[]].

/ Marcus Comstedt (ACROSS) (Hail Ilpalazzo!)

The test will fail, I guess.

If that's a problem, one could hack the server program to try random ports until bind succeeds, and then write the port number to stdout.

/ Niels Möller (ny flexiblare bröstkorg)

Previous text:

2003-04-15 23:46: Subject: Starting new pike processes from the testsuite

---

What happens is port 25678 is already in use? (For instance, by another testsuite)

/ Fredrik (Naranek) Hubinette (Real Build Master)

Why not just do bind(0) and report which port was bound?

/ Fredrik (Naranek) Hubinette (Real Build Master)

Previous text:

2003-04-15 23:48: Subject: Starting new pike processes from the testsuite

---

The test will fail, I guess.

If that's a problem, one could hack the server program to try random ports until bind succeeds, and then write the port number to stdout.

/ Niels Möller (ny flexiblare bröstkorg)

The testsuite for the Unicode module might also prove enlighting.

/ Martin Nilsson (har bott i google)

Previous text:

2003-04-15 11:42: Subject: Starting new pike processes from the testsuite

---

Excerpts from the main testsuite.in:

test_do([[ // This is a memleak test rm("testsuite_test.pike"); Stdio.write_file("testsuite_test.pike", "constant Q=10000000000000000;\n" "int main() {}\n"); Process.system(RUNPIKE +" testsuite_test.pike"); rm("testsuite_test.pike"); ]])

test_any([[object p=Process.create_process(Process.split_quoted_string(RUNPIKE)+({"-e","exit(2)"})); sleep(10); return p->wait()]],2)

/ Henrik Grubbström (Lysator)

I don't know what the interface should be like. There already is SSL.sslport, which mimics Stdio.Port. The https.pike program is just a dummy server with a hardcoded port, certificate, and key, used for testing.

/ Niels Möller (ny flexiblare bröstkorg)

Previous text:

2003-04-15 17:34: Subject: Starting new pike processes from the testsuite

---

If there is HTTPs server code, it probably belongs - would be great - in the Protocols.HTTP.Server tree.

/ Mirar

Sure, redesigning the interface of the SSL code would be nice, but not something I'm going to do any time soon.

All I want for now is some tests to help exersize the reorganization of Crypto...

/ Niels Möller (ny flexiblare bröstkorg)

Previous text:

2003-04-15 17:38: Subject: Starting new pike processes from the testsuite

---

How about a SSL.Port class for a Stdio.Port emulator?

/ Per Hedbor ()

The problems is that when *changes* are made to the Crypto API, however trivial (for example renamed methods or classes), the SSL code must be changed too, I want some SSL testcases to help with that.

/ Niels Möller (ny flexiblare bröstkorg)

Previous text:

2003-04-15 17:49: Subject: Starting new pike processes from the testsuite

---

I think the best way to go about is to extend the testsuite of the Crypto (and _Crypto) module to test all API:s.

/ Martin Nilsson (har bott i google)

I read the masters thesis about it a few months ago, and I think it explained quite clearly how the module works. But then I guess I'm a little biased ;-)

/ Niels Möller (ny flexiblare bröstkorg)

Previous text:

2003-04-15 18:09: Subject: Starting new pike processes from the testsuite

---

The testsuite already ensures that all modules compiles. That aside, I have myself done one or two lame attempts to make SSL tests, but it's not like the module is well documented or easy to understand.

/ Martin Nilsson (har bott i google)

That sounds like it should be really easy to implement just inheriting/mimicking HTTP.Server.Port and just changing create and the Stdio.Port part.

The tricky parts is certificates and stuff and API for that, I guess.

/ Mirar

Previous text:

2003-04-15 17:37: Subject: Starting new pike processes from the testsuite

---

I don't know what the interface should be like. There already is SSL.sslport, which mimics Stdio.Port. The https.pike program is just a dummy server with a hardcoded port, certificate, and key, used for testing.

/ Niels Möller (ny flexiblare bröstkorg)