1 Notes on Lisp implementations for Maxima:
2 =========================================
4 Clisp, CMUCL, Scieneer Common Lisp (SCL), GCL (ANSI-enabled only),
5 ECL, ABCL and SBCL can compile and execute Maxima.
6 Allegro Common Lisp (ACL) and CCL might also work, but have not
9 Ports to other ANSI Common Lisps should be straightforward
10 and are welcome; please post a message on the Maxima mailing list
11 if you are interested in working on a port.
13 When Maxima is recompiled, the Lisp implementation is selected by
14 an argument of the form `--enable-foolisp` for the configure script.
15 `./configure --help` shows a list of the Lisp types recognized by
16 configure (among other options). It is possible to specify several
17 Lisp type(s) you want Maxima to be built with at the same time.
18 configure tries to autodetect the Lisp type if it is not specified,
19 but it has been reported that autodetection can fail.
22 Comparison of execution times (in seconds)
23 ------------------------------------------
25 This is the result of the `run_testsuite()` function for
26 Maxima 5.36.0 as reported on the Maxima mailing list:
28 ### 64 Bit (Gentoo Linux):
36 ### 32 Bit (Gentoo Linux)
46 Clisp <http://clisp.org>
47 ------------------------
49 Clisp can be built with readline support, so Maxima has
50 advanced command-line editing facilities when built with it.
52 Clisp is compiled to bytecodes, so Maxima running on Clisp is
53 substantially slower than on Lisps compiled to machine instructions.
54 On the other hand, Clisp contains code from CLN <https://www.ginac.de/CLN/>,
55 a library for efficient computations with all kinds of numbers in
56 arbitrary precision. Another advantage of Clisp is that byte code
57 resulting in compiling a program on one computer might work on a
58 computer running a different Clisp version. Also Clisp uses an
59 extremely efficient memory handling which means it might not run
60 out of memory where ECL, SBCL and GCL do.
62 CLISP version 2.49 has a bug that causes it to output garbled data
63 if the front-end is fast enough to acknowledge a data packet while
64 the next data packet is still being prepared.
66 There are Clisp implementations for many platforms including
67 MS Windows and Unix-like systems.
69 Maxima compiled with a typical linux install of clisp 2.49.92
70 typically depends on the following libraries:
80 CMUCL <https://cmucl.org/>
81 ----------------------------------
83 CMUCL is a fast option for Maxima on platforms where it is
84 available. The rmaxima front-end provides advanced line-editing
85 facilities for Maxima when compiled with CMUCL. rlwrap is available
86 from: <https://github.com/hanslub42/rlwrap>
88 CMUCL versions: 18e and 19a and later are known to work.
90 There are CMUCL implementations only for Unix-like systems
93 Scieneer Common Lisp (SCL) <https://web.archive.org/web/20171014210404/http://www.scieneer.com/scl/>
94 ----------------------------------------------------------------------------------------------------
96 Scieneer Common Lisp (SCL) is a fast option for Maxima for a
97 range of Linux and Unix platforms. The SCL 1.2.8 release and later
98 are supported. SCL offers a lower case, case sensitive, version which
99 avoids the Maxima case inversion issues with symbol names. Tested
100 front end options are: Maxima emacs mode available in the
101 interfaces/emacs/ directory, the Emacs imaxima mode available from
102 <https://sites.google.com/site/imaximaimath/>, and TeXmacs available from
103 <https://www.texmacs.org>
105 GCL <https://www.gnu.org/software/gcl/>
106 ---------------------------------------
108 GCL versions starting with 2.4.3 can be built with readline
109 support, so Maxima has advanced command-line editing facilities
110 when built with it. GCL produces a fast Maxima executable that
111 profit from GCL's fast bignum algorithms.
113 Only the ANSI-enabled version of GCL works with Maxima, i.e.,
114 when GCL is built, it must be configured with the `--enable-ansi` flag,
115 i.e., execute `./configure --enable-ansi` in the build directory
116 before executing make.
118 Whether GCL is ANSI-enabled or not can be determined by
119 inspecting the banner which is printed when GCL is executed;
120 if ANSI-enabled, the banner should say `ANSI`.
121 Also, the special variable `*FEATURES*` should include the keyword `:ANSI-CL`.
123 There are GCL implementations for many platforms
124 including MS Windows and Unix-like systems.
126 Maxima compiled using a typical linux install using gcl 2.6.12
127 typically depends on:
136 SBCL <https://www.sbcl.org>
137 --------------------------
139 SBCL is a fork of CMUCL which differs in some minor details,
140 but most notably, it is simpler to rebuild SBCL than CMUCL.
141 For many tasks a Maxima compiled with SBCL is considerably faster than
142 GCL. For other tasks GCL is faster than SBCL.
144 As SBCL doesn't use readline it is recommended to use rmaxima for using
145 a command-line Maxima with SBCL. For common details of SBCL and CMUCL
148 Maxima compiled using a typical Linux install using SBCL 1.4.10
149 typically depends on:
155 Allegro Common Lisp <https://franz.com/products/allegro-common-lisp/>
156 ---------------------------------------------------------------------
158 Maxima should work with Allegro Common Lisp, but
159 only limited testing has been done with these Lisp
160 implementations. User feedback would be welcome.
163 CCL <https://ccl.clozure.com/>
164 ------------------------------
166 CCL, formerly known as OpenMCL, is known to work with Maxima on
167 all platforms where ccl runs including Linux, Mac OSX, and Windows.
168 There are appear to be some bugs in the 32-bit version of CCL, but
169 the 64-bit version passes all tests.
172 ECL <https://common-lisp.net/project/ecl/>
173 ------------------------------------------
175 ECL is known to work with Maxima and passes the testsuite. ECL
176 runs on many platforms and OSes and is the lisp compiler used for
177 Maxima on Android. It is faster than CLISP, but seems to tend more
178 towards fragmenting the available memory. ECL tends to be slower
179 than GCL or SBCL but faster than CLISP.
181 ECL must be configured to use the C compiler, building Maxima with the
182 ECL bytecode compiler is (currently) not possible. So do **not** use the
183 option `--with-cmp=no` when building ECL.
185 Maxima compiled using a typical linux install using ecl 16.1.2
186 typically only depends on:
191 Armed Bear Common Lisp (ABCL) <https://www.abcl.org>
192 ----------------------------------------------------
194 ABCL's main feature is that is tightly integrated into Java.
195 That also means that it is an interpreter running in a virtual machine
196 which makes it even slower than Clisp. Also Java doesn't automatically
197 convert tail-recursive function calls to loops which means that in a
198 few functions might run out of stack space faster than other Lisps.