Mailman 3 September 2016 - pypy-dev

About gurobipy
by Luis José Novoa 31 May '18

31 May '18

Hi All, My name is Luis Novoa. Im working on a large scale optimization problem using python/gurobipy for my implementation. I was wondering if any of you in this list has used gurobipy with pypy, or if it is even possible at this time. Thanks in advance, Luis.

9 9

improve error message when missing 'self' in method definition
by Carl Friedrich Bolz 04 Oct '16

04 Oct '16

Hi all, I read this paper today about common mistakes that Python beginners make: https://www.researchgate.net/publication/307088989_Some_Trouble_with_Transp… The most common one by far is forgetting the "self" parameter in the method definition (which also still happens to me regularly). The error message is not particularly enlightening, if you don't quite understand the explicit self in Python. So I wonder whether we should print a better error message, something like this: $ cat m.py class A(object): def f(x): return self.x A().f(1) $ pypy m.py Traceback (application-level): File "m.py", line 4 in <module> A().f(1) TypeError: f() takes exactly 1 argument (2 given). Did you forget 'self' in the function definition? It's a bit the question how clever we would like this to be to reduce false positives, see the attached patch for a very simple approach. Anyone have opinions? Cheers, Carl Friedrich

4 6

Sorting structs
by Tuom Larsen 04 Oct '16

04 Oct '16

Dear list! I stumbled upon a problem and I was wondering if someone would by so kind and explain to me what is going on. The problem is sorting 2 000 000 points by `x` coordinate: from random import random from time import time class point(object): def __init__(self, x, y): self.x, self.y = x, y data = [point(random(), random()) for i in range(2000000)] t = time() data.sort(key=lambda p:p.x) print time() - t on my machine in runs 8.74s under PyPy 5.4.1 on MacOS. I then try to sort the points in JavaScript: var data = []; for (var i=0; i<2000000; i++) { data.push({x:Math.random(), y:Math.random()}); } console.time('sorting'); data.sort(function(a, b) { return a.x - b.x; }); console.timeEnd('sorting'); and it runs in 3.09s under V8 5.3. I was just wondering, why is it nearly 3x slower under PyPy than it is under V8? Is there any way I could make the code run faster? Thank you in advance!

3 4

NumPy or NumPyPy
by Papa, Florin 28 Sep '16

28 Sep '16

Hello, I read this announcement [1] saying that "over 99% of the upstream numpy test suite" is passed. Is this when using pypy with the upstream numpy (thanks to the incremental improvements brought to cpyext) or is it when using pypy with numpypy? I also found this link [2], tracking numpypy status. Is numpypy still maintained or does this buildbot track the regression tests for upstream numpy in pypy (using cpyext)? [1] https://morepypy.blogspot.com/2016/08/pypy2-v54-released-incremental.html [2] http://buildbot.pypy.org/numpy-status/latest.html Thank you, Florin

3 2

Change how a pypy executable finds its lib-python/lib_pypy directories?
by Armin Rigo 26 Sep '16

26 Sep '16

Hi all, Right now the "pypy" executable looks for its lib-python/lib_pypy directories by starting from the path of the "pypy" executable and going up. That path is derived from the C-level argv[0], on which we try once to call readlink() if it is a symlink. That logic is in pypy.module.sys.initpath. When using cffi's embedding mode, however, the logic starts differently: we start at the path of the "libpypy-c.so" library instead, and then we look around in the same way. This is because, of course, there is no "pypy" executable in this case. This difference breaks some packaged pypy's: it is the only thing that prevents e.g. the Arch Linux pypy from working in embedded mode. The reason is that they install /opt/pypy/* like we recommend, and set a symlink /usr/bin/pypy -> /opt/pypy/bin/pypy like we also recommend, but they really move libpypy-c.so in /usr/lib/libpypy-c.so. It works fine in the regular case, but fails with the embedding mode. I tried to make a symlink /usr/lib/libpypy-c.so -> /opt/pypy/bin/libpypy-c.so and embedding mode then works fine. As an attempt to fix this more generally, we should make the two cases more similar. Does it make sense to only look around from the "libpypy-c.so" location? Finding the path of "libpypy-c.so" requires a little bit more hackery than just reading "argv[0]", but on the other hand it is not likely to break even if "pypy" is called strangely (because argv[0] can in theory be any random string). It is code that is so far in pypy.module._cffi_backend.embedding, which would probably move to pypy.module.sys.initpath. This change would require minor fixes for pypy packagers---e.g. arch linux would simply make a symlink for /usr/lib/libpypy-c.so instead of for /usr/bin/pypy. Makes sense? A bientôt, Armin.

2 5

PyPy 3.3? (and: informal Cape Town sprint)
by Armin Rigo 23 Sep '16

23 Sep '16

Hi Ronan, hi all, Can we fix a date after which we stop needing to update pypy 3.3? It is a (minor) mess that, whenever I find a bug or small missing feature in pypy 3.5, I usually try first to see if it should be done in the py3k (3.3) branch or directly in py3.5. In fact I would guess by now that I have misplaced one or two such fixes. Sorry about that. It would help reduce the overhead if we could all focus on py3.5. At the start of August, Ronan mentioned the last pypy 3.3 release being 1 or 2 months away. Ronan, can you say what you'd still like to do and what you'd like fixed first for the release? I can help you with that. I suggest that we tentatively fix the date of October the 8th for the release (this is just after PyCon ZA). Then we can then focus our South African sprint on py3.5. About the sprint: anyone around Cape Town for the two weeks after PyCon ZA is welcome to join. We don't have exactly a plan so far; for sure we will not be sprinting without some break days. A bientôt, Armin.

4 7

Re: [pypy-dev] pypy_win64_support
by Armin Rigo 20 Sep '16

20 Sep '16

Hi Diana, Answering you on the pypy-dev mailing list. The question is about Win64 support (http://doc.pypy.org/en/latest/windows.html#what-is-missing-for-a-full-64-bi…): On 19 September 2016 at 14:34, Popa, Diana <diana.popa(a)intel.com> wrote: > First of all, I started by translating PyPy on win64 with a hacked 2.7 Python as described in the provided URL. I am encountering a problem when trying to calculate MAX_DIGITS_THAT_CAN_FIT_IN_INT = rbigint.fromint(-sys.maxint - 1).numdigits(). (...) The URL doesn't describe what you are doing: it tells about hacking Python 2.7, yes, but the next thing it says you should do is trying to pass a specific subset of the tests. There is no point in immediately trying to translating PyPy. But anyway, if you tried doing that, you would hit the same problem immediately. This is caused by the "precision loss" I wrote about on that page. I assume that sys.maxint is equal to 2**63-1 in your hacked Python. Then likely some operations occur in C on this "long long" value that temporarily cast the value to "long", resulting in precision loss. You have to track down what it is. For example, start by writing some tests and put them in rpython.rlib.test.test_win64: def test_basic_arithmetic(): def eq(x, y): assert type(x) is type(y) is int return x == y x = 2**50 assert eq(x, 2**50) assert str(x) == "1125899906842624" assert str(sys.maxint) == "9223372036854775807" assert eq(sys.maxint, int(2**63-1)) assert eq(x + x, 2**51) assert eq(x - (-x), 2**51) assert eq(x * 4, 2**52) assert eq(x | x, 2**50) assert eq(x >> 1, 2**49) etc. ...as I guess that there is code a bit everywhere that depends on that. Fixing such failures is a matter of tracking down where CPython stores things temporarily inside "long". It is not necessary to get a perfect CPython that handles "long long" systematically, but at least the basics like above should work. Alternatively, you could try to globally replace "long" with "long long" and similar. A bientôt, Armin.

1 0

RevDB released
by Armin Rigo 10 Sep '16

10 Sep '16

Hi all, It is my pleasure to announce the first beta release of RevDB: a "reverse debugger" for Python. A reverse debugger is a debugger where you can go forward and backward in time. RevDB is designed to track down the annoying, hard-to-reproduce bug in your Python program. It is based on PyPy, but it is useful even if you usually run your programs on CPython. Note that Win32 is not supported so far. As usual here, this is Python 2.7 only, but it should not be too hard to add support for revdb in the py3k/py3.5 branches. https://bitbucket.org/pypy/revdb/ A bientôt, Armin

1 0

Build Pypy with different Interpreters
by Jan Brohl 09 Sep '16

09 Sep '16

Is it possible to build different interpreters like Stackless, IronPython or Jython? If not - why? If yes - is it (in theory) possible to gain a speedup on those without GIL? (Is there multithreading at all the in translation process?) Thanks, Jan

3 5

Any known tricks or effort made to make the PyPy source code build faster?
by Wang, Peter Xihong 08 Sep '16

08 Sep '16

Hi, By default, it appears most of the time during the build/compile process, only 1 single CPU core is busy, signaling missing of parallel compiling. Is there any best known practice to make it faster? Thanks, Peter

4 7