Get PARI/GP to stop starting automatically

[kcrisman]

I still am getting the known behavior below when exiting Sage.

Exiting Sage (CPU time 0m0.74s, Wall time 11m11.95s). 
Exiting spawned GP/PARI interpreter process. 

It's not exactly a bug, but it's also annoying and potentially confusing. We should stop it.

Apply 9959_combined.patch below and install the new spkg: ​http://sage.math.washington.edu/home/jdemeyer/spkg/pari-2.4.3.svn-12577.p9.spkg (note that this spkg is not based on the earlier p8)

[leif]

Oh, I thought you just meant "GP/PARI" should read "PARI/GP"...

[leif]

So how do we manage this?

Rewrite the class sage.interfaces.gp.Gp to lazily start the interpreter, or initialize sage.interfaces.gp.gp to None and explicitly do gp = Gp() everywhere it is used unless gp != None?

[jdemeyer]

I believe this in Gp.__init__ is the culprit:

# gp starts up with this set to 1, which makes pexpect hang: 
self.set_default('breakloop',0)

If this is the case, an easy solution is to add a patch to the pari spkg to set breakloop to 0 by default.

[jdemeyer]

Replying to leif:

Oh, I thought you just meant "GP/PARI" should read "PARI/GP"...

This should also be changed.

[leif]

Replying to jdemeyer:

I believe this in Gp.__init__ is the culprit:

# gp starts up with this set to 1, which makes pexpect hang: 
self.set_default('breakloop',0)

Good catch. I read this a dozen times without noticing that this actually calls gp... 8/

If this is the case, an easy solution is to add a patch to the pari spkg to set breakloop to 0 by default.

Try it...

[jdemeyer]

Renames "GP/PARI" to "PARI/GP"

[jdemeyer]

New spkg for testing: ​http://sage.math.washington.edu/home/jdemeyer/spkg/pari-2.4.3.svn-12577.p8.spkg

[leif]

Wow, a lot of places where "GP/PARI" had to be changed... (well done)

We have two similar instances in the scripts repo:

local/bin/sage-sage:    echo "  -gp <..>      -- run Sage's GP (Pari) with given arguments"
local/bin/sage-sage.py:                     help="run Sage's GP (Pari), passing it all remaining arguments")

But I think we don't need to change these on this ticket.

Patches look fine, will test them later (with the new spkg of course).

[leif]

The spkg's changelog entry (and patches/README.txt) could describe the reason for patching default.c a bit more. (The ticket number is only in the commit message.)

[leif]

s|and|and/or|

[cremona]

OK, so I was the one who put in those two breakloop calls, which I only did since otherwise things did not work properly -- but in the testing I just did I found it only necessary to delete the brealoop call in __init__, while keeping the one in _start. Are you both sure that it is safe to delete the one in _start as well? Does that not cause gp to enter the breakloop on encountering an error, which is not what you want?

If you are right (and I assumed that you did test this!) then I am wondering what else has changed, since when we started the pari upgrade it was definitely necessary to deal with this.

And by the way, the *only* reason I created a _start function for Gp different from the default one in PExpect was to add this line in _start; so if that line is not needed, you can probably delete the while _start function for class Gp.

[leif]

I haven't come to testing yet... (including a closer look at the relevant PARI sources).

It should at least not hurt to keep the second set_default('breakloop',0) in Gp._start().

[leif]

Replying to leif:

It should at least not hurt to keep the second set_default('breakloop',0) in Gp._start().

As far as I can see, we don't need it there, since gp will of course (re)start with the hard-coded defaults whenever _start() is called.

(This is different to setting it "manually" from the Expect interface each time gp is [re]started.)

So we can IMHO drop Gp._start() completely, as John noted.

[jdemeyer]

Replying to cremona:

Are you both sure that it is safe to delete the one in _start as well? Does that not cause gp to enter the breakloop on encountering an error, which is not what you want?

The new spkg patches the gp source code to disable the breakloop by default, so we can remove all references to breakloop from Sage. Note also that the patch 9959_gp_error_doctest.patch​ actually acts a doctest to check this.

[jdemeyer]

Don't set the 'breakloop' default in Sage

[jdemeyer]

Add doctest for error recovery

[jdemeyer]

Replying to cremona:

And by the way, the *only* reason I created a _start function for Gp different from the default one in PExpect was to add this line in _start; so if that line is not needed, you can probably delete the while _start function for class Gp.

Done

[cremona]

It's a pity that we had to patch the gp source code -- clearly that was an option right from the start, and what I did was to avoid that. Best (in my opinion) would be if gp had a command-line option to turn off the breakloop default, since Sage could easily use that. It might be worth suggesting that possibility upstream.

[jdemeyer]

Patch for the PARI spkg .p7 to .p8 (for review)

[jdemeyer]

Replying to cremona:

It's a pity that we had to patch the gp source code -- clearly that was an option right from the start, and what I did was to avoid that. Best (in my opinion) would be if gp had a command-line option to turn off the breakloop default, since Sage could easily use that. It might be worth suggesting that possibility upstream.

If one would add an option to gp, add an option for non-interactive (script) use which would also disable breakloop by default. I remember I proposed this (a long time ago) to the PARI/GP developers without much success.

[cremona]

Replying to jdemeyer:

Replying to cremona:

It's a pity that we had to patch the gp source code -- clearly that was an option right from the start, and what I did was to avoid that. Best (in my opinion) would be if gp had a command-line option to turn off the breakloop default, since Sage could easily use that. It might be worth suggesting that possibility upstream.

If one would add an option to gp, add an option for non-interactive (script) use which would also disable breakloop by default. I remember I proposed this (a long time ago) to the PARI/GP developers without much success.

I suppose that from their point of view it makes no sense to use gp non-interactively, since gp is the interactive interface to the pari library!

[jdemeyer]

Alternative solution:

When gp starts, it reads a configuration file (by default, $HOME/.gprc). If the environment variable $GPRC is set, it uses that as a filename for the .gprc file. We could create a file $HOME/.sage/gp/gprc similarly to $HOME/.sage/ipyhton/ipythonrc and have Sage set $GPRC to that location.

Then, the file $HOME/.sage/gp/gprc should contain a line:

breakloop =  0

[jdemeyer]

Replying to cremona:

I suppose that from their point of view it makes no sense to use gp non-interactively, since gp is the interactive interface to the pari library!

Good point :-)

[leif]

Replying to jdemeyer:

Alternative solution:

When gp starts, it reads a configuration file (by default, $HOME/.gprc). If the environment variable $GPRC is set, it uses that as a filename for the .gprc file. We could create a file $HOME/.sage/gp/gprc similarly to $HOME/.sage/ipyhton/ipythonrc and have Sage set $GPRC to that location.

Then, the file $HOME/.sage/gp/gprc should contain a line:

breakloop =  0

I was just going to suggest that, too.

Note that we should have two of them, one for the interactive gp provided by Sage, and one for the pexpect interface.

[jdemeyer]

Replying to leif:

Note that we should have two of them, one for the interactive gp provided by Sage, and one for the pexpect interface.

I think we need only one, for the pexpect interface. I would prefer sage -gp to use my $HOME/.gprc.

Setting this to needs_work to implement the alternative solution.

[leif]

Replying to jdemeyer:

If one would add an option to gp, add an option for non-interactive (script) use which would also disable breakloop by default. I remember I proposed this (a long time ago) to the PARI/GP developers without much success.

Perhaps they meanwhile chnaged their minds.. ;-)

There's some stuff in it for using gp from Emacs or TeXmacs?, but only as a compile time option IIRC. Also some kind of "non-interactive" gp use.

[leif]

Replying to jdemeyer:

Replying to leif:

Note that we should have two of them, one for the interactive gp provided by Sage, and one for the pexpect interface.

I think we need only one, for the pexpect interface. I would prefer sage -gp to use my $HOME/.gprc.

Right, but we should take care not to use the other one (i.e. by setting GPRC and therefore overriding $HOME/.gprc) if we start the interactive gp.

[jdemeyer]

Anybody happens to know where is the code to deal with $DOT_SAGE, to populate it with files?

[jdemeyer]

Use $SAGE_ROOT/local/etc/gprc.expect as .gprc file

[jdemeyer]

New spkg: ​http://sage.math.washington.edu/home/jdemeyer/spkg/pari-2.4.3.svn-12577.p9.spkg. This is based on p7 (not p8) and adds the gprc.expect file. So installing the new spkg and applying all 4 patches should do it.

[jdemeyer]

Doesn't work yet, stupid mistake.

[jdemeyer]

sage.math.washington.edu seems much slower than usual today...

[leif]

Does gp expand $SAGE_ROOT in the gprc.expect file, or does Sage do that before it starts gp?

[jdemeyer]

Replying to leif:

Does gp expand $SAGE_ROOT in the gprc.expect file, or does Sage do that before it starts gp?

gp does that.

[jdemeyer]

.gprc file for the Gp() interface

[jdemeyer]

Patch for the PARI spkg .p7 to .p9 (for review)

[jdemeyer]

Fixed spkg: ​http://sage.math.washington.edu/home/jdemeyer/spkg/pari-2.4.3.svn-12577.p9.spkg

[jdemeyer]

All 4 sagelib patches combined

[cremona]

The sagelib patch looks good to me.

For review, please could you give instructions for the simple-minded as to how to apply the patch which converts .p7 to .p9? And then how to rebuild? Thanks.

Of course, someone else who knows how to do this reliably is welcome to get in first with a review.

[jdemeyer]

Replying to cremona:

The sagelib patch looks good to me.

For review, please could you give instructions for the simple-minded as to how to apply the patch which converts .p7 to .p9?

Well, the easiest is to install the new spkg:

sage -i http://sage.math.washington.edu/home/jdemeyer/spkg/pari-2.4.3.svn-12577.p9.spkg

And then how to rebuild?

I personally do not know the most reliable way to rebuild a complete Sage installation. One thing which works for sure is the following:

• download sage-4.6.alpha3.tar and extract it so we have a clean, uncompiled sage-4.6.alpha3.
• rm spkg/standard/pari-*.spkg
• download [ttp://sage.math.washington.edu/home/jdemeyer/spkg/pari-2.4.3.svn-12577.p9.spkg] and and put it in spkg/standard
• now make Sage as usual

[cremona]

THanks -- I am doing that, will do a complete new build (with the sagelib patch too of course) and report back.

[cremona]

I made a freshly unpacked 4.6.alpha3, and replaced its pari spkg with the p9 version; then built all Sage; then applied the sagelib patch on this ticket; then tested the whole library.

All passed, so here's a positive review.