Use shorter float repr when possible

[noam]

BPO	1580
Nosy	@gvanrossum, @tim-one, @nascheme, @rhettinger, @amauryfa, @mdickinson, @pitrou, @ericvsmith, @tiran, @avassalotti, @ericsnowcurrently
Files	short_float_repr.diff ieee_dbl.c float.diff: Tentative patch for py3k float2.diff: Improved tentative patch for py3k unnamed unnamed

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = <Date 2009-04-17.11:19:21.231>
created_at = <Date 2007-12-10.19:13:26.972>
labels = ['interpreter-core', 'type-feature']
title = 'Use shorter float repr when possible'
updated_at = <Date 2014-03-07.21:30:36.354>
user = 'https://bugs.python.org/noam'

bugs.python.org fields:

activity = <Date 2014-03-07.21:30:36.354>
actor = 'eric.snow'
assignee = 'none'
closed = True
closed_date = <Date 2009-04-17.11:19:21.231>
closer = 'mark.dickinson'
components = ['Interpreter Core']
creation = <Date 2007-12-10.19:13:26.972>
creator = 'noam'
dependencies = []
files = ['8910', '8920', '10808', '10840', '13176', '13199']
hgrepos = []
issue_num = 1580
keywords = ['patch']
message_count = 109.0
messages = ['58359', '58364', '58374', '58394', '58398', '58399', '58401', '58403', '58405', '58407', '58411', '58416', '58419', '58422', '58424', '58427', '58429', '58430', '58432', '58433', '58436', '58443', '58445', '58449', '58452', '58454', '58456', '58457', '58466', '58467', '58473', '58474', '58477', '58478', '58480', '58509', '58510', '58524', '58707', '58710', '58726', '58727', '58728', '58731', '58733', '58736', '58748', '58753', '58787', '58790', '58966', '59053', '59054', '69232', '69235', '69242', '69243', '69368', '69421', '69422', '69429', '69552', '69554', '82708', '82710', '82711', '82718', '82724', '82727', '82728', '82730', '82752', '82774', '82776', '82777', '82801', '82807', '82817', '82832', '82994', '82995', '83049', '83050', '83189', '83199', '83200', '84666', '85690', '85692', '85693', '85696', '85697', '85698', '85705', '85733', '85739', '85741', '85747', '86048', '86069', '86127', '86129', '86131', '86141', '86144', '86146', '86149', '86158', '117723']
nosy_count = 15.0
nosy_names = ['gvanrossum', 'tim.peters', 'nascheme', 'rhettinger', 'amaury.forgeotdarc', 'mark.dickinson', 'pitrou', 'eric.smith', 'christian.heimes', 'alexandre.vassalotti', 'noam', 'jaredgrubb', 'preston', 'olau', 'eric.snow']
pr_nums = []
priority = 'normal'
resolution = 'accepted'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'enhancement'
url = 'https://bugs.python.org/issue1580'
versions = ['Python 3.1']

[noam]

The current float repr() always calculates the 17 first digits of the
decimal representation of a float, and displays all of them (discarding
trailing zeros). This is a simple method for making sure that
eval(repr(f)) == f. However, many times it results in a long string,
where a shorter string would have sufficed. For example, currently
repr(1.1) is '1.1000000000000001', where obviously '1.1' would have been
good enough, and much easier to read.

This patch implements an algorithm for finding the shortest string that
will evaluate to the right number. It is based on the code from
http://www.cs.indiana.edu/~burger/fp/index.html, and also on the
floating-point formatting function of TCL, Tcl_PrintDouble.

The patch also adds a test case, which takes a long list of floating
point numbers, created especially for testing binary to decimal
conversion, and makes sure that eval(repr(f)) == f. See
floating_points.txt for the source of the list.

[gvanrossum]

I like this; but I don't have time for a complete thourough review.
Maybe Tim can lend a hand?

If Tim has no time, I propose that if it works correctly without leaks
on at least Windows, OSX and Linux, we check it in, and worry about more
review later.

Crys, can you test this on Windows and see if it needs any project file
updates? I'll check OSX. Linux works and I see no leaks.

[tiran]

Applied in r59457

I had to add checks for _M_X64 and _M_IA64 to doubledigits.c. The rest
was fine on Windows.

[gvanrossum]

Noam, perhaps you can help with this? We checked this in but found a
problem: repr(1e5) suddenly returns '1.0'. Can you think of a cause for
this?

[noam]

I don't know, for me it works fine, even after downloading a fresh SVN
copy. On what platform does it happen?

[tiran]

I've disabled the new repr() in trunk and py3k until somebody has sorted
out the build problems. I've removed doubledigits.c from Makefile.in and
disabled the code with #ifdef Py_BROKEN_REPR in floatobject.c.

[gvanrossum]

On what platform does it happen?

Linux on x86.

It seems find on PPC OSX.

This suggests it could be a byte order bug.

[noam]

I also use linux on x86. I think that byte order would cause different
results (the repr of a random float shouldn't be "1.0".)
Does the test case run ok? Because if it does, it's really strange.

[tim-one]

There is nothing you can do to repr() that's sufficient by itself to
ensure eval(repr(x)) == x.

The other necessary part is correctly rounded float /input/ routines.

The 754 standard does not require correctly rounding input or output
routines. It does require that eval(repr(x)) == x provided that repr(x)
produce at least 17 significant digits (not necessarily correctly
rounded, but "good enough" so that eval(repr(x)) == x). That's why
Python produces 17 significant digits. While there's no guarantee that
all platform I/O routines are good enough to meet the 754 requirement,
most do, and what Python's repr() actually does is the strongest that
/can/ be done assuming no more than full 754 conformance.

Again, this cannot be improved cross-platform without Python supplying
both output and input routines. Burger's output routines supply only
half of what's needed. An efficient way to do the other half (correctly
rounding input) was first written up by Clinger. Note that neither the
Burger/Dybvig output code nor the Clinger input code are used much in
practice anymore; David Gay's code is usually used instead for "go fast"
reasons:

http://mail.python.org/pipermail/python-list/2004-July/272167.html

Clinger also endorses Gay's code:

ftp://ftp.ccs.neu.edu/pub/people/will/retrospective.pdf

However, as the Python list link says, Gay's code is "mind-numbingly
complicated".

There is no easy cross-platform solution, where "cross-platform" means
just "limited to 754 platforms".