Version 0.2, January 18, 2011
Albert Graef <Dr.Graef@t-online.de>
A Pure interface to GNU Octave.
This is an Octave module for the Pure programming language, based on Paul Kienzle’s octave_embed which allows Octave to be embedded in other languages. It allows you to execute arbitrary Octave commands and Octave functions from Pure.
pure-octave is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.
pure-octave is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
Please see the accompanying COPYING file for the precise license terms. The GPL can also be read online at http://www.gnu.org/licenses/.
Get the latest source from http://pure-lang.googlecode.com/files/pure-octave-0.2.tar.gz.
Run make to compile the module and make install (as root) to install it in the Pure library directory. This requires GNU make, and of course you need to have both Pure and Octave installed (including Octave’s mkoctfile utility and the corresponding header files and libraries).
make tries to guess your Pure installation directory and platform-specific setup. If it gets this wrong, you can set some variables manually. In particular, make install prefix=/usr sets the installation prefix. Please see the Makefile for details.
NOTE: This release of pure-gen has been tested with Octave 3.2.4. Older versions might require some fiddling with the sources to get the embedded Octave interface working.
Import this module into your Pure scripts as follows:
using octave;
This will add an embedded instance of the Octave interpreter to your program. (You can import this module as often as you want, but there’s always only one instance of Octave in each process.)
Executes arbitrary Octave code.
> octave_eval "eig([1 2;3 4])";
ans =
-0.37228
5.37228
0
This prints the result on stdout and returns a result code (zero if everything went fine). To suppress the printing of the result, simply terminate the Octave statement with a semicolon:
> octave_eval "eig([1 2;3 4]);";
0
This allows you to define values to be used when evaluating Octave code, and to transfer results back to Pure. However, before such globals can be accessed in Octave, you must explicitly declare them as globals:
> octave_eval "global x y ans";
0
Now you can use octave_set and octave_get to transfer values between Pure and Octave as follows:
> octave_set "x" {1.0,2.0;3.0,4.0};
{1.0,2.0;3.0,4.0}
> octave_eval "eig(x);";
0
> octave_get "ans";
{-0.372281323269014;5.37228132326901}
Note that the most recent result can always be accessed through Octave’s ans variable. You can also use an explicit variable definition as follows:
> octave_eval "y = eig(x);";
0
> octave_get "y";
{-0.372281323269014;5.37228132326901}
Call an octave function in a direct fashion. fun denotes the name of the function, n the number of function results and args the function arguments.
> let x = {1.0,2.0;3.0,4.0};
> octave_call "eig" 1 x;
{-0.372281323269014;5.37228132326901}
Note the second argument, which denotes the desired number of return values. This will usually be 1 (or 0 if you don’t care about the result), but some Octave functions may return a variable number of results, depending on how they’re called. Multiple values are returned as tuples in Pure:
> octave_call "eig" 2 x;
{-0.824564840132394,-0.415973557919284;0.565767464968992,-0.909376709132124},
{-0.372281323269014,0.0;0.0,5.37228132326901}
If there are multiple arguments, you can specify them either as a tuple or a list:
> octave_call "norm" 1 (x, 2);
5.46498570421904
> octave_call "norm" 1 [x, 1];
6.0
Instead of a function name, you can also specify the function to be called using a special kind of Octave object, a function value. These are returned, e.g., by Octave’s str2func and inline builtins. For your convenience, pure-octave provides a frontend to these builtins, the octave_func function, which lets you specify an Octave function in one of two ways:
Returns the Octave function with the given name. This works like Octave’s str2func builtin.
Returns an “inline” function, where expr is an Octave expression (as a string) describing the function value. This works like Octave’s inline builtin. Instead of just an Octave expression, you can also specify a tuple or a list consisting of the inline expression and the parameter names. (Otherwise the parameters are determined automatically, see the description of the inline function in the Octave manual for details.)
Note that inline functions allow you to call stuff that is not an Octave function and hence cannot be specified directly in octave_call, such as an operator. Examples:
> let eig = octave_func "eig";
> let mul = octave_func "x*y";
> octave_call eig 1 (octave_call mul 1 (x,x));
{0.138593383654928;28.8614066163451}
> let add = octave_func ("x+y","x","y");
> octave_call add 1 (x,x);
{2.0,4.0;6.0,8.0}
As shown above, the octave_set, octave_get and octave_call functions convert Pure data to corresponding Octave values and vice versa. Octave scalars and matrices of boolean, integer, double, complex and character data are all supported by this interface, and are mapped to the corresponding Pure data types in a straightforward manner (scalars to scalars, matrices to matrices and strings to strings). Note that in any case these conversions create copies of the data, so modifying, say, an Octave matrix received via octave_get in Pure only affects the Pure copy of the matrix and leaves the original Octave matrix unchanged.
Any other kind of Octave object (including Octave function objects, see Direct Function Calls) is just passed through as is, in the form of a cooked pointer to an Octave value which frees itself when garbage-collected. You can check for such objects with the octave_valuep predicate:
Check for Octave value pointers.
> let eig = octave_func "eig";
> eig; octave_valuep eig;
#<pointer 0x230dba0>
1
Such Octave value pointers can be used freely whereever an Octave value is needed (i.e., in invocations of octave_set and octave_call).
You should also note the following:
Octave’s cell arrays and structures are not supported at this time, so they will show up as opaque Octave value pointers in Pure land. This allows these objects to be passed around freely, but you can’t inspect or modify them in Pure.
Scalars and 1x1 matrices are indistinguishable in Octave, thus any 1x1 matrix will be returned as a scalar from Octave to Pure.
All types of boolean and integer matrices are returned from Octave to Pure as (machine) integer matrices. When converted back to Octave, these all become Octave int32 matrices, but you can easily convert them to boolean or other types of matrices in Octave as needed. For instance:
> octave_set "a" {1,2;3,4};
{1,2;3,4}
> octave_eval "global a ans";
0
> octave_eval "eig(a)";
error: eig: wrong type argument `int32 matrix'
1
> octave_eval "eig(double(a))";
ans =
-0.37228
5.37228
0
> octave_eval "a>2";
ans =
0 0
1 1
0
> octave_get "ans";
{0,0;1,1}
Octave strings are mapped to Pure strings, and character matrices with more than one row are mapped to (symbolic) column vectors of Pure strings. Example:
> octave_set "a" "Hello, world!";
"Hello, world!"
> octave_eval "a";
a = Hello, world!
0
> octave_eval "[a;'abc']";
ans =
Hello, world!
abc
0
> octave_get "ans";
{"Hello, world!";"abc "}
The embedded Octave interpreter provides one special builtin, the pure_call function which can be used to call any function defined in the executing Pure script from Octave. For instance:
> even m::matrix = {~(int x mod 2) | x=m};
> octave_eval "pure_call('even', 1:12)";
ans =
0 1 0 1 0 1 0 1 0 1 0 1
0
Here’s the description of the pure_call function, as it is printed with Octave’s help command:
`pure_call' is a built-in function
RES = pure_call(NAME, ARG, ...)
[RES, ...] = pure_call(NAME, ARG, ...)
Execute the Pure function named NAME (a string) with the given arguments.
Arguments and result types may be scalars and matrices of boolean, integer,
double, complex and character data. The Pure function may return multiple
results as a tuple. Example: pure('succ', 99) => 100.
Directly embedding Octave in Pure programs is convenient because it allows easy exchange of data between Pure and Octave, and you can also call Octave functions directly from Pure and vice versa. However, it also comes at a cost. A default build of Octave pulls in quite a few dependencies of its own which might conflict with other modules loaded in a Pure script. Specifically, we have found that building Octave with native graphics support (as is included in the latest Octave versions) may give problems with third-party graphics libraries such as VTK (which segfaults, apparently due to FreeType library compatibility issues, if used in the same program as Octave).
We’re not sure whether Octave or VTK is to blame here, but until these quirks get ironed out you can work around them by disabling native graphics support in Octave. Unfortunately, Octave currently doesn’t provide an option to do so, but the following patches available at the Pure website will do the trick (there are different versions of the patch, corresponding to the current stable and development versions of Octave):
The patch is to be applied to Octave’s configure.in (configure.ac for the latest development versions) by running a command like the following in the Octave source directory:
patch -p0 < octave-3.2.4.patch
Then just rerun autoconf and configure and compile Octave as usual. The resulting version of Octave will have native graphics disabled and should work ok with third party libraries such as VTK.
(If you have an Octave version which differs considerably from what we provide here then you may have to rework the patch accordingly or apply the necessary changes to configure.in or configure.ac manually.)