%timeit on command line doesn't recognize [1..10] syntax

[jasongrout]

sage: %time [1..10]
------------------------------------------------------------
   File "<timed exec>", line 1
     [1..10]
          ^
<type 'exceptions.SyntaxError'>: invalid syntax

sage: %timeit [1..10]
------------------------------------------------------------
   File "<magic-timeit>", line 6
     [1..10]
          ^
<type 'exceptions.SyntaxError'>: invalid syntax

sage: %timeit xrange(11)
1000000 loops, best of 3: 392 ns per loop

Component: user interface

Keywords: timeit ipython

Author: Mike Hansen

Reviewer: Jason Grout

Merged: sage-4.3.1.rc1

Issue created by migration from https://trac.sagemath.org/ticket/797

[jasongrout]

comment:3

The workaround now is to use the timeit command, which does send things to the preparser.

sage: timeit('[1..10]')
625 loops, best of 3: 220 µs per loop

[williamstein]

comment:5

Just because there is a workaround doesn't mean this should be closed. I reported this problem to Fernando Perez of Ipython and he "admitted" that it is a bug, and was working on fixing it at that last sage days. I'm re-opening this.

By the way, it's generally very bad for anybody but me or Michael Abshoff to close tickets on trac!

[jasongrout]

comment:6

I only closed it after mabshoff gave his okay on IRC. I guess I should have posted the IRC log.

[simon-king-jena]

Replying to @jasongrout:

Here is some account of what happens in Sage 3.2.3:

sage: %time [1..10]
CPU times: user 0.00 s, sys: 0.00 s, total: 0.00 s
Wall time: 0.00 s
[1, 2, 3, 4, 5, 6, 7, 8, 9, 10]

So, the first problem is not a problem anymore.

sage: %timeit [1..10]
------------------------------------------------------------
   File "<magic-timeit>", line 6
     [1..10]
          ^
SyntaxError: invalid syntax

So, that is the only remaining issue. I am now looking at sage/misc/sage_timeit.py.

[mwhansen]

comment:8

We can overwrite the default magic commands in IPython:

sage: _ip.expose_magic("timeit", lambda self, s: timeit(s))
sage: %timeit [1..10]
625 loops, best of 3: 58.4 µs per loop

[simon-king-jena]

comment:9

I somehow got the impression that it is an issue with preparsing, for the following reason:

sage: timeit('[1..10]',preparse=False)
------------------------------------------------------------
   File "<magic-timeit>", line 6
     [1..10]
          ^
SyntaxError: invalid syntax

This is the same error as in

sage: %timeit [1..10]

In the code, by default preparsing should be done; namely, in sage_timer.py it says

def sage_timeit(stmt, globals, preparse=None, number = 0, repeat = 3, precision = 3):
...
    if preparse is None:
        preparse = interpreter.do_preparse

Now, we also have

sage: from sage.misc import interpreter
sage: interpreter.do_preparse
True

So, it should be with preparsing.

For testing, I inserted a line in the code that prints the value of preparse.

The curious thing is that this line is executed when doing timeit('[1..10]') or timeit.eval('[1..10]'), but it is not executed when doing %timeit [1..10].

I was just told that %timeit ... should do the same as timeit.eval('...'), but apparently it doesn't. So, perhaps it is a problem with the %?

Can you tell me where the behaviour of % is defined?

[mwhansen]

comment:10

Are you doing this in the notebook or the command-line? They're two totally separate things.

Can you hop on IRC?

[mwhansen]

comment:11

When you run %timeit from the command-line, it runs code in IPython and does not touch any code in the Sage library. The issue is that the IPython magic command "%timeit" doesn't do the preparsing. Fernando was working on an upstream fix for this. An easy downstream fix for this would be to do as I suggested above and overwrite the IPython version of timeit with our own with the expose_magic function.

sage: _ip.expose_magic("timeit", lambda self, s: timeit(s))
sage: %timeit [1..10]
625 loops, best of 3: 58.4 µs per loop

The right place to put this would probably be in local/bin/ipy_profile_sage.py.

[simon-king-jena]

comment:12

Replying to @mwhansen:

When you run %timeit from the command-line, it runs code in IPython and does not touch any code in the Sage library. The issue is that the IPython magic command "%timeit" doesn't do the preparsing. Fernando was working on an upstream fix for this. An easy downstream fix for this would be to do as I suggested above and overwrite the IPython version of timeit with our own with the expose_magic function.

sage: _ip.expose_magic("timeit", lambda self, s: timeit(s))
sage: %timeit [1..10]
625 loops, best of 3: 58.4 µs per loop

The right place to put this would probably be in local/bin/ipy_profile_sage.py.

Thank you!

While you replied, I tried to post the following:

Replying to @mwhansen:

Are you doing this in the notebook or the command-line? They're two totally separate things.

It is command-line, and I don't know what happens on the notebook. Can you test it?

If it is in preparser_ipython, then the following might be the source of trouble:

def preparse_ipython(line, reset=True):
    global num_lines
    global q_lines

    L = line.lstrip()
    if L.startswith('%'):
        # This should be installed as an Ipython magic command,
        # but I don't know how yet...
        L = L[1:].strip()
        import sage.interfaces.all
        if L.lower() in sage.interfaces.all.interfaces:
            switch_interface(L.lower())
            return "''"
        else:
            # only preparse non-magic lines
            return line

Hence, if the line starts with % and if the word after % is not the name of an interface than simply the line is returned unchanged.

But what happens afterwards? In the end of the day, %timeit gets an interpretation and does call some timeit command!

Can you hop on IRC?

I try (I never used IRC before). So, if it works, see you soon.

Cheers
Simon

[simon-king-jena]

Attachment: timeit.patch.gz

Change preparse_ipython so that %timeit results in calling timeit.eval

[simon-king-jena]

comment:13

In preparse_ipython, lines starting with % have not been changed unless they refer to some interface. I added support for %timeit.

Idea: Call timeit.eval

With the patch, the following examples now work:

sage: %timeit [1..10]
625 loops, best of 3: 127 µs per loop
sage: %timeit 'a'=="a"
625 loops, best of 3: 285 ns per loop

and the other ways of calling the timer still work:

sage: %time [1..10]
CPU times: user 0.00 s, sys: 0.00 s, total: 0.00 s
sage: timeit('[1..10]')
625 loops, best of 3: 127 µs per loop

[simon-king-jena]

Changed keywords from none to timeit ipython

[simon-king-jena]

comment:14

Replying to @simon-king-jena:

In preparse_ipython, lines starting with % have not been changed unless they refer to some interface. I added support for %timeit.

Idea: Call timeit.eval

Probably I should elaborate more on how it works.

In the old version of preparse_ipython, it was tested whether an input line starts with '%'. If this was the case, then the word after '%' was checked. If that word referred to an interface, the interface was used. If the word after '%' was differently, the line was returned without a change.

For the new version, I suggest to test whether the word after '%' is 'timeit'. If it is, preparse_ipython returns timeit.eval(..rest of the input line..).

[simon-king-jena]

comment:15

To emphasize another detail: My suggestion only relates with %timeit on the command line, not on the notebook.

[jasongrout]

comment:16

This patch works for me, but I think mhansen ought to comment on this approach versus the approach he advocated above using _ip.expose_magic("timeit", lambda self, s: timeit(s))

[jasongrout]

comment:17

The _ip.expose_magic approach also seems to work just as well. I would prefer using the ipython mechanism for doing this, rather than complicating the preparser.

In fact, I'd suggest using ipython to do all of the %mode functionality as well, if it's a possibility.

[jasongrout]

comment:18

Simon,

what do you think about the idea of using ipython to do this, like mhansen suggests above? I think it's cleaner, it keeps the ipython stuff together (i.e., the "%" commands with ipython), and avoids making the preparser more complicated.

[simon-king-jena]

comment:19

Hi Jason,

Replying to @jasongrout:

what do you think about the idea of using ipython to do this, like mhansen suggests above? I think it's cleaner, it keeps the ipython stuff together (i.e., the "%" commands with ipython), and avoids making the preparser more complicated.

Sure, it makes sense. The problem is that i have no idea about ipython, so, I guess i would not be able to do it in that way. Perhaps i should look at local/bin/ipy_profile_sage.py, but i wouldn't be upset if someone else were faster... :)

Cheers,
Simon

[jasongrout]

comment:20

Changing this to "needs work" in light of mhansen's solution above.

[mwhansen]

Attachment: trac_797.patch.gz

[mwhansen]

Author: Mike Hansen

[mwhansen]

comment:21

I've put a patch up which implements the expose_magic idea.

[jasongrout]

comment:22

Looks great! Positive review to Mike's patch. Only apply Mike's patch (to the local/bin repository!)

[rlmill]

Merged: sage-4.3.1.rc1

[rlmill]

Reviewer: Jason Grout