can't exit or detach after error in attached file

[mstreng]

17:57:03:~$ touch myfile.sage
17:57:05:~$ sage
----------------------------------------------------------------------
| Sage Version 5.8, Release Date: 2013-03-15                         |
| Type "notebook()" for the browser-based notebook interface.        |
| Type "help()" for help.                                            |
----------------------------------------------------------------------
sage: %attach myfile.sage
sage: file('myfile.sage', 'w').write('1/0')
sage: 1+1
...
ZeroDivisionError: Rational division by zero
sage: exit
...
ZeroDivisionError: Rational division by zero
sage: detach('myfile.sage')
...
ZeroDivisionError: Rational division by zero

It seems like the only way to continue using this Sage session is to fix the error first. But suppose you have some reason not to (e.g. are in a hurry, and the error is hard to fix). I think it would be much better if Sage was to evaluate "1+1" and especially "exit" and "detach" even when unable to reload one of the attached files.

Possible ways to fix this are discussed on sage-devel.

Apply

• attachment: trac_14523_improve_attach.patch
• attachment: trac_14523_readline.patch

Depends on #14266

CC: @nexttime @nbruin @seblabbe

Component: user interface

Author: Volker Braun

Reviewer: Travis Scrimshaw

Merged: sage-5.11.rc0

Issue created by migration from https://trac.sagemath.org/ticket/14523

[vbraun]

comment:1

• only reload once per time stamp change
• reload without having to press any key in Sage
• collect all attach-related files in one module instead of sprinkling it around

[vbraun]

Author: Volker Braun

[hughrthomas]

comment:3

I'm not knowledgeable enough to offer a useful review of the code. It does seem like a good solution to the observed problem.

There is a behaviour of "attach" which bugs me. If the file you try to attach generates an error when you attach it, then it doesn't actually get attached. I don't see the point of this. It makes even less sense now than before. Would it be easy to fix on this ticket, or should I open another one?

I do also have one comment about the current patch -- if it wouldn't be too complicated, I think it would be nice to get a new sage prompt after the reloading is finished, to indicate that it has completed.

[vbraun]

comment:4

Updated patch

• remember that files is attached even if load fails
• don't reload files that are in the process of being written

[hughrthomas]

comment:6

Thanks very much! (I have tested attaching a file with an error.)

[hughrthomas]

comment:7

Bonus: this also seems to fix the issue reported at #14149: when files are attached, spurious files are created in the current directory.

[vbraun]

comment:8

Updated patch fixes some minor things, makes imports lazy, and adds doctests.

[hughrthomas]

comment:9

@Volker: Thanks for the explanation (at #14149), and the new doctests, which clarify what is happening with the creation of files. Would it be possible to create the temp files somewhere else? (I guess there are security concerns with putting them in /tmp, but isn't there any alternative?) Or, would it be possible to keep track of them and remove them? I guess at each update, the previous copy could be deleted, and detach or quit could delete the otherwise-remaining copy?

[vbraun]

comment:10

I think the "debug mode" for load and attach should be removed. Files should always be preparsed in memory. We should keep a user-accessible list of loaded/attached files with some state, preparsed form, etc. Any debugging that you can do with temporary files can be done much nicer by querying the files on the command line.

But thats for another ticket.