Running gap with incorrect GAP_ROOT_PATHS results in awful error messages

[tornaria]

• For gap expect interface:

$ GAP_ROOT_PATHS=/bad sage -c 'gap(1)'
** Gap crashed or quit executing 'SaveWorkspace("/home/tornaria/.sage/gap/tmp1xkouu8e");' **
Restarting Gap and trying again
** Gap crashed or quit executing '\$sage1:=1;;' **
Restarting Gap and trying again
** Gap crashed or quit executing 'SaveWorkspace("/home/tornaria/.sage/gap/tmpq435w0q8");' **
Restarting Gap and trying again
Traceback (most recent call last):
  File "/usr/lib/python3.12/site-packages/sage/interfaces/expect.py", line 545, in _start
    self._expect.expect(self._prompt)
  File "/usr/lib/python3.12/site-packages/pexpect/spawnbase.py", line 354, in expect
    return self.expect_list(compiled_pattern_list,
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib/python3.12/site-packages/pexpect/spawnbase.py", line 383, in expect_list
    return exp.expect_loop(timeout)
           ^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib/python3.12/site-packages/pexpect/expect.py", line 179, in expect_loop
    return self.eof(e)
           ^^^^^^^^^^^
  File "/usr/lib/python3.12/site-packages/pexpect/expect.py", line 122, in eof
    raise exc
pexpect.exceptions.EOF: End Of File (EOF). Exception style platform.
Gap with PID 7218 running /usr/bin/gap -A -l /bad -b -p -T -E -m 64m /usr/lib/python3.12/site-packages/sage/ext_data/gap/sage.g
command: /usr/bin/gap
args: ['/usr/bin/gap', '-A', '-l', '/bad', '-b', '-p', '-T', '-E', '-m', '64m', '/usr/lib/python3.12/site-packages/sage/ext_data/gap/sage.g']
buffer (last 100 chars): b''
before (last 100 chars): b"'-l <gaproot>'?@J@n If you ran the GAP binary directly, try running the 'gap.sh' or 'gap.bat' scr\\@J"
after: <class 'pexpect.exceptions.EOF'>
match: None
match_index: None
exitstatus: None
flag_eof: True
pid: 7218
child_fd: 8
closed: False
timeout: None
delimiter: <class 'pexpect.exceptions.EOF'>
logfile: None
logfile_read: None
logfile_send: None
maxread: 4194304
ignorecase: False
searchwindowsize: None
delaybeforesend: None
delayafterclose: 0.1
delayafterterminate: 0.1
searcher: searcher_re:
    0: re.compile(b'gap> ')

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/usr/lib/python3.12/site-packages/sage/interfaces/expect.py", line 1520, in __init__
    self._name = parent._create(value, name=name)
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib/python3.12/site-packages/sage/interfaces/interface.py", line 517, in _create
    self.set(name, value)
  File "/usr/lib/python3.12/site-packages/sage/interfaces/gap.py", line 1353, in set
    self._eval_line(cmd, allow_use_file=True)
  File "/usr/lib/python3.12/site-packages/sage/interfaces/gap.py", line 728, in _eval_line
    self._start()
  File "/usr/lib/python3.12/site-packages/sage/interfaces/gap.py", line 1187, in _start
    gap_reset_workspace(verbose=False)
  File "/usr/lib/python3.12/site-packages/sage/interfaces/gap.py", line 1526, in gap_reset_workspace
    g.save_workspace()
  File "/usr/lib/python3.12/site-packages/sage/interfaces/gap.py", line 1283, in save_workspace
    self.eval('SaveWorkspace("%s");' % (f.name), allow_use_file=False)
  File "/usr/lib/python3.12/site-packages/sage/interfaces/gap.py", line 526, in eval
    result = Expect.eval(self, input_line, **kwds)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib/python3.12/site-packages/sage/interfaces/expect.py", line 1414, in eval
    return '\n'.join(self._eval_line(L, allow_use_file=allow_use_file, **kwds)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib/python3.12/site-packages/sage/interfaces/expect.py", line 1414, in <genexpr>
    return '\n'.join(self._eval_line(L, allow_use_file=allow_use_file, **kwds)
                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib/python3.12/site-packages/sage/interfaces/gap.py", line 728, in _eval_line
    self._start()
  File "/usr/lib/python3.12/site-packages/sage/interfaces/gap.py", line 1192, in _start
    Expect._start(self, "Failed to start GAP.")
  File "/usr/lib/python3.12/site-packages/sage/interfaces/expect.py", line 549, in _start
    raise RuntimeError("unable to start %s: %s" % (self.name(), msg))
RuntimeError: unable to start gap: End Of File (EOF). Exception style platform.
Gap with PID 7218 running /usr/bin/gap -A -l /bad -b -p -T -E -m 64m /usr/lib/python3.12/site-packages/sage/ext_data/gap/sage.g
command: /usr/bin/gap
args: ['/usr/bin/gap', '-A', '-l', '/bad', '-b', '-p', '-T', '-E', '-m', '64m', '/usr/lib/python3.12/site-packages/sage/ext_data/gap/sage.g']
buffer (last 100 chars): b''
before (last 100 chars): b"'-l <gaproot>'?@J@n If you ran the GAP binary directly, try running the 'gap.sh' or 'gap.bat' scr\\@J"
after: <class 'pexpect.exceptions.EOF'>
match: None
match_index: None
exitstatus: None
flag_eof: True
pid: 7218
child_fd: 8
closed: False
timeout: None
delimiter: <class 'pexpect.exceptions.EOF'>
logfile: None
logfile_read: None
logfile_send: None
maxread: 4194304
ignorecase: False
searchwindowsize: None
delaybeforesend: None
delayafterclose: 0.1
delayafterterminate: 0.1
searcher: searcher_re:
    0: re.compile(b'gap> ')

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/usr/lib/sagemath/10.3.beta7/bin/sage-eval", line 10, in <module>
    eval(compile(s,'<cmdline>','exec'))
  File "<cmdline>", line 1, in <module>
  File "/usr/lib/python3.12/site-packages/sage/interfaces/interface.py", line 306, in __call__
    result = self._coerce_from_special_method(x)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib/python3.12/site-packages/sage/interfaces/interface.py", line 334, in _coerce_from_special_method
    return (x.__getattribute__(s))(self)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "sage/structure/sage_object.pyx", line 749, in sage.structure.sage_object.SageObject._gap_ (build/cythonized/sage/structure/sage_object.c:8742)
  File "sage/structure/sage_object.pyx", line 725, in sage.structure.sage_object.SageObject._interface_ (build/cythonized/sage/structure/sage_object.c:8165)
  File "/usr/lib/python3.12/site-packages/sage/interfaces/interface.py", line 299, in __call__
    return cls(self, x, name=name)
           ^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib/python3.12/site-packages/sage/interfaces/expect.py", line 1525, in __init__
    raise TypeError(*x.args)
TypeError: unable to start gap: End Of File (EOF). Exception style platform.
Gap with PID 7218 running /usr/bin/gap -A -l /bad -b -p -T -E -m 64m /usr/lib/python3.12/site-packages/sage/ext_data/gap/sage.g
command: /usr/bin/gap
args: ['/usr/bin/gap', '-A', '-l', '/bad', '-b', '-p', '-T', '-E', '-m', '64m', '/usr/lib/python3.12/site-packages/sage/ext_data/gap/sage.g']
buffer (last 100 chars): b''
before (last 100 chars): b"'-l <gaproot>'?@J@n If you ran the GAP binary directly, try running the 'gap.sh' or 'gap.bat' scr\\@J"
after: <class 'pexpect.exceptions.EOF'>
match: None
match_index: None
exitstatus: None
flag_eof: True
pid: 7218
child_fd: 8
closed: False
timeout: None
delimiter: <class 'pexpect.exceptions.EOF'>
logfile: None
logfile_read: None
logfile_send: None
maxread: 4194304
ignorecase: False
searchwindowsize: None
delaybeforesend: None
delayafterclose: 0.1
delayafterterminate: 0.1
searcher: searcher_re:
    0: re.compile(b'gap> ')

• For libgap:

$ GAP_ROOT_PATHS=/bad sage -c 'libgap(1)'
gap: hmm, I cannot find 'lib/init.g' maybe use option '-l <gaproot>'?
------------------------------------------------------------------------
/usr/lib/python3.12/site-packages/cysignals/signals.so(+0x9820)[0x7f6abd4cf820]
/usr/lib/python3.12/site-packages/cysignals/signals.so(+0x98e6)[0x7f6abd4cf8e6]
/usr/lib/python3.12/site-packages/cysignals/signals.so(+0xc320)[0x7f6abd4d2320]
/usr/lib/libc.so.6(+0x3d910)[0x7f6abd63d910]
/usr/lib/libgap.so.8(DoOperation1Args+0x24)[0x7f6a2b6ce644]
/usr/lib/libgap.so.8(GAP_EvalString+0x5c)[0x7f6a2b69ab2c]
/usr/lib/python3.12/site-packages/sage/libs/gap/util.so(+0xcbf4)[0x7f6a2c675bf4]
/usr/lib/python3.12/site-packages/sage/libs/gap/libgap.so(+0x1839a)[0x7f6a2c60639a]
/usr/lib/python3.12/site-packages/sage/structure/coerce_maps.so(+0x1de70)[0x7f6ab7dcbe70]
/usr/lib/python3.12/site-packages/sage/structure/parent.so(+0x285ae)[0x7f6abc1795ae]
/usr/lib/python3.12/site-packages/sage/misc/lazy_import.so(+0x1673d)[0x7f6abc62873d]
/usr/lib/libpython3.12.so.1.0(_PyObject_MakeTpCall+0x6b)[0x7f6abdc006eb]
/usr/lib/libpython3.12.so.1.0(+0x1804ea)[0x7f6abdb804ea]
/usr/lib/libpython3.12.so.1.0(PyEval_EvalCode+0xb2)[0x7f6abdcbf752]
/usr/lib/libpython3.12.so.1.0(+0x17e364)[0x7f6abdb7e364]
/usr/lib/libpython3.12.so.1.0(+0x21f903)[0x7f6abdc1f903]
/usr/lib/libpython3.12.so.1.0(PyObject_Vectorcall+0x4f)[0x7f6abdc00c0f]
/usr/lib/libpython3.12.so.1.0(+0x1804ea)[0x7f6abdb804ea]
/usr/lib/libpython3.12.so.1.0(PyEval_EvalCode+0xb2)[0x7f6abdcbf752]
/usr/lib/libpython3.12.so.1.0(+0x2ddca7)[0x7f6abdcddca7]
/usr/lib/libpython3.12.so.1.0(+0x2ddc1b)[0x7f6abdcddc1b]
/usr/lib/libpython3.12.so.1.0(+0x2de302)[0x7f6abdcde302]
/usr/lib/libpython3.12.so.1.0(_PyRun_SimpleFileObject+0x1cf)[0x7f6abdcde08f]
/usr/lib/libpython3.12.so.1.0(_PyRun_AnyFileObject+0x44)[0x7f6abdcdddd4]
/usr/lib/libpython3.12.so.1.0(Py_RunMain+0x313)[0x7f6abdce66f3]
/usr/lib/libpython3.12.so.1.0(Py_BytesMain+0x37)[0x7f6abdce6287]
/usr/lib/libc.so.6(+0x27c4c)[0x7f6abd627c4c]
/usr/lib/libc.so.6(__libc_start_main+0x85)[0x7f6abd627d05]
/usr/bin/python3(_start+0x21)[0x5620698a0081]
------------------------------------------------------------------------
...

[tornaria]

@orlitzky

Here's a situation where GAP_ROOT_PATHS will be incorrect: if one installs sagemath-standard in a venv using system gap the default value for SAGE_LOCAL will be sys.prefix which is the venv, so SAGE_LOCAL/share/gap will not be /usr/share/gap. Thus causing this issue.

So I see three issues related to this:

1. make it so that an error initializing gap or libgap is actually detected and raise a reasonable error instead of keep going and fail in a confusing way.
2. fix the default of GAP_ROOT_PATHS so it will be good when SAGE_LOCAL is different than the system prefix (/usr).
3. do we actually need GAP_ROOT_PATHS? It seems to me that running gap without giving gap root paths should be more correct, at least for running gap / libgap.

[orlitzky]

3. do we actually need GAP_ROOT_PATHS? It seems to me that running gap without giving gap root paths should be more correct, at least for running gap / libgap.

It's "needed" so that people can install the gap_packages SPKG and use them from the system gap. The SAGE_LOCAL path is appended to the usual system one so that GAP will look there for packages, too.

If you're not using sage-the-distro, the easiest fix is just to set GAP_ROOT_PATHS to the correct value, i.e. the one system directory. But if you want to start a bunch of fights then yeah it would be cleaner to not support sage-the-distro GAP packages when using the system GAP. Naturally I think everyone should be using the system packages instead of the SPKGs.

[tornaria]

3. do we actually need GAP_ROOT_PATHS? It seems to me that running gap without giving gap root paths should be more correct, at least for running gap / libgap.

It's "needed" so that people can install the gap_packages SPKG and use them from the system gap. The SAGE_LOCAL path is appended to the usual system one so that GAP will look there for packages, too.

Ah, thanks, I missed this. Did you mean "the SAGE_LOCAL path is prepended to the usual system one"? (so that pkgs installed in sage can override system gap pkgs).

Let me see if I get this right: what we need is a configuration, let's tentatively call it SAGE_GAP_PKG which when set will be prepended to the system gap root paths. Is that correct?

This means:

a. for gap binary, we can run using -l "$SAGE_GAP_PKG;" where the empty last component means to prepend to system root paths:

$ gap -r -l "/SAGE/GAP/PATH;" -q -c 'Display(KERNEL_INFO().GAP_ROOT_PATHS);QUIT;'
[ "/SAGE/GAP/PATH/", "/usr/lib64/gap/", "/usr/share/gap/" ]

b. for libgap we can run using -l "$SAGE_GAP_PKG;<system-root-paths>":

$ ./libgap -r -l "/SAGE/GAP/PATH;/usr/lib64/gap/;/usr/share/gap/" -q -c 'Display(KERNEL_INFO().GAP_ROOT_PATHS);QUIT;'
[ "/SAGE/GAP/PATH/", "/usr/lib64/gap/", "/usr/share/gap/" ]

(here ./libgap is just a trivial shim to run gap-as-library)

c. the for libgap we can get from the gap binary:

$ gap -r --systemfile <(echo 'PRINT_TO("*stdout*",KERNEL_INFO().GAP_ROOT_PATHS,"\n");QuitGap();')
[ "/usr/lib64/gap/", "/usr/share/gap/" ]

(with the systemfile trick this is quick enough to do at runtime)

If this sounds right, I can try to adjust #37344.

If you're not using sage-the-distro, the easiest fix is just to set GAP_ROOT_PATHS to the correct value, i.e. the one system directory. But if you want to start a bunch of fights then yeah it would be cleaner to not support sage-the-distro GAP packages when using the system GAP. Naturally I think everyone should be using the system packages instead of the SPKGs.

I don't want to fight, and I don't want to improve sagemath-standard at the expense of sage-the-distro.

But I want sane defaults and avoid any unnecessary configuration. The less knobs we have, the easier it is to run in different situations. As a matter of fact, the problem I describe here first arised when I built sage-the-distro in editable mode. Because sage-conf editable mode is BROKEN so it won't load GAP_ROOT_PATHS.

A different situation is when running sagemath-standard in a venv since SAGE_LOCAL/share/gap is wrong (SAGE_LOCAL defaults to sys.prefix which is /usr when running non-isolated but path-to-venv when running in a venv).

[orlitzky]

That sounds more or less right for now but the whole thing is a mess so I'm sure some use case will wind up broken. IMO what we really need is a sagelib build system ASAP so we don't need sage_conf handed to us by a deity before sagelib will work.