The Mercury compiler has a backend that generates Java source code, that can then be compiled into Java bytecode suitable for running using the Java SE runtime system. The backend is mostly complete, but some parts of the Mercury standard library are not yet implemented.
The Java backend requires Java SE 8 or higher -- older versions of Java are not supported.
- Prerequisites
- Installing the
java
grade - Compiling programs with the
java
grade - Running programs with the
java
grade - Limitations
- Library support
- Interfacing with Java
- Performance
- Path length restrictions on Windows
- Java compiler memory exhaustion
- Mercury-level debugging
- Java-level debugging
- Building the Mercury compiler in the
java
grade
In order to use Mercury's Java backend you will need:
-
The Java SE platform.
-
OR another compatible Java implementation, such as OpenJDK or Amazon Coretto.
The Mercury compiler uses the grade java
to target Java source code that is
then compiled into Java bytecode by the Java compiler.
Mercury's autoconfiguration script will cause the java
grade to be installed
if it finds a suitable Java compiler (e.g. javac
) and Java runtime
(e.g. java
) in your PATH
.
You can check if your Mercury installation has been configured to include the
java
grade by looking if java
is included in the output of the Mercury
compiler's --output-stdlib-grades
option.
Once you have a Mercury installation that includes the java
grade, you can
build programs such as hello.m
or calculator.m
in the samples
directory.
$ cd samples
$ mmc --grade java --make hello
When building programs with the java
grade you must use mmc --make
; using
mmake
to build programs using the java
grade is not supported.
You can run the hello
program from the previous section by doing:
$ ./hello
Note that hello
is a simple shell script generated by the Mercury compiler
that invokes the program using the Java interpreter. The actual class files are
packaged up into a Java archive (JAR) named hello.jar
.
If you are using the Windows command-line interpreter, i.e. cmd.exe
, then
setting the value of the option --target-env-type
to "windows" will cause the
Mercury compiler to generate a batch file that invokes the program, instead of
a shell script.
(See the "Using the Mercury compiler" section of the The Mercury User's Guide for further details.)
The following features of the Mercury implementation are not (currently) supported by the Java backend:
- Mercury-level debugging (however, see further down).
- Mercury-level profiling.
- Trailing.
- Tabling.
- Backjumping.
The Mercury standard library has not been fully ported to Java yet. The use of unimplemented procedures will result in a run-time error, with a stack trace and a message like:
Sorry, not implemented: foreign code for this function
If you find missing functionality, you can interface to Java using Mercury's foreign language interface.
The following individual Mercury standard library procedures are either not supported or not fully implemented:
-
io.read_binary/{3,4}
io.write_binary/{3,4}
The current implementation of
read_binary
does not work with the way Mercury file streams are implemented for the Java backend. -
benchmarking.report_stats/0
benchmarking.report_full_memory_stats/0
Memory usage statistics are not yet available, and cpu time is not the same as in the C backends, as per
time.m
. -
io.environment.set_environment_var/{4,5}
The Java APIs do not support setting environment variables, hence this predicate throws an exception.
-
store.arg_ref/5
store.new_arg_ref/5
Due to some limits in RTTI support, dynamic type checking is missing for these predicates. They should be used with care.
-
time.clock/3
time.clocks_per_sec/0
time.times/7
time.clk_tck/0
Because the Java APIs do not provide a way of implementing these procedures exactly in pure Java, we have approximated them with what is available.
-
math.fma/3
This function is not available because it is not supported by Java 8. (It will be supported once the minimum version of Java required by Mercury increases.)
You can call Java code directly from Mercury using the foreign language interface. For example:
:- pred to_string(T::in, string::out) is det.
:- pragma foreign_proc("Java",
to_string(T::in, Str::out),
[promise_pure, will_not_call_mercury],
"
Str = T.toString();
").
The implementation will include this Java code in the module's .java file.
You can then call the predicate to_string/2
exactly the same as if it were
implemented using pure Mercury code.
For more information about the foreign language interface, see the Mercury Language Reference Manual. Additionally, the samples/java_interface directory in the Mercury distribution contains examples of how to use the foreign language interface with Java.
Short programs may run much more slowly in the java
grade than the C grades.
The runtime is probably dominated by Java class loading and running in
interpreted mode. A long running program should perform reasonably well with a
Just-In-Time compiler. It may also be possible to use an Ahead-Of-Time Java
compiler, but we have not tried that yet.
When using the java
grade on Windows, it is sometimes possible for the fully
qualified names of generated files to exceed the maximum path length. If this
occurs the Mercury compiler will abort with a message like:
Uncaught Mercury exception:
Software Error: parse_tree.module_cmds: predicate \
`parse_tree.module_cmds.list_class_files_for_jar'/6: \
Unexpected: io.file_type failed: No such file or directory
In this case all that can (currently) be done is to reduce the length of the
build path, for example by shifting the build directory closer to the root of
the file system (e.g. C:\mercury
).
It is possible for the Java compiler to run out of memory when compiling
Java code generated by the Mercury compiler. If that occurs, you can pass
an option to the javac
program to increase the limit. For example:
$ mmc --make foo --java --java-flag -J-Xmx512m
Or pass an option to javac
using a Mercury.options
file:
JAVACFLAGS += -J-Xmx512m
The only Mercury-level debugger available for the Java backend is the experimental source-to-source debugger; see README.ssdebug.md for details.
By default, javac
already generates line number and source file debugging
information. You can include local variable debugging information by specifying
--target-debug
when invoking the Mercury compiler. For example:
$ mmc --grade java --target-debug --make <progname>
You can then use the jdb
debugging tool, which comes as part of the Java SDK
distribution, to debug your program. For more information, see the
documentation for javac
and jdb
.
Building the Mercury compiler and other related tools in the Java grade is NOT generally supported and should be considered experimental. In particular, a Mercury compiler built in the Java grade may be slower than normal and some features may not be available.
However, if you want to give it a try, the required steps are:
-
Ensure that you have an existing working Mercury compiler in your
PATH
and a clean version of the Mercury source tree. -
Run
prepare.sh
andconfigure
as normal. -
Add the line:
GRADE=java
to a file named
Mmake.params
at the top-level of the source tree. -
Begin the build process using the following command:
$ mmake --use-mmc-make GRADE=java
The Java version of the compiler MUST be built using
mmake
's--use-mmc-make
option; the build will not work otherwise. Setting the variableGRADE
in the invocation ofmmake
is currently necessary in order to avoid variable definition ordering problems inMmake.workspace
. -
To install the Java version of the compiler, do:
$ mmake --use-mmc-make install GRADE=java