3 nitc - compiles Nit programs.
8 nitc [*options*] FILE...
13 nitc is the current official Nit compiler.
14 It takes the main module of a Nit program as argument and produces an executable file.
16 By default, the generated executables are produced in the current directory.
17 (see `--dir` for details.)
19 Internally, nitc rely on the presence of a C compiler. Usually gcc (but nitc was successfully tested with clang).
20 A compilation directory is therefore created and (re-)used.
21 By default, the compilation directory is named `nit_compile` and is removed after the compilation.
22 (see `--compile-dir` for details.)
24 Currently, because Nit is still in heavy development, the compilation directory is not cleaned after the compilation.
26 By default, the compilation process tries to have a good trade-off between the compilation time and the performance of produced executables.
27 To produce more optimized executables, the current best option is `--semi-global`.
29 To improve the compilation time and simplify the compilation of multiple programs, more than one file can be given.
30 Each one will be compiled into a distinct executable.
32 $ nitc prog1.nit prog2.nit
34 To combine files into a single program, use the `-m` option.
36 $ nitc prog1.nit -m other_module.nit
38 nitc can produces executables for various platforms when specific modules are used.
39 Currently, android, pnacl and emscripten are supported.
40 See the documentation of these specific modules for details.
48 Show additional warnings (advices).
50 By default, only important warnings are displayed.
51 May be overridden by `-w`.
53 Important warnings are displayed by default. A warning is considered important when:
55 * There is a simple correction.
56 * There is no reason to let the code this way.
57 * There is always a real issue (no false positive).
59 Other warnings, called advices, are not displayed by default to avoid filling the terminal with
61 A warning is considered an advice when:
63 * The correction could be complex. e.g. require a refactorisation or an API change.
64 * The correction cannot be done. e.g. Code that use a deprecated API for some compatibility reason.
65 * There is not a real issue (false positive). Note that this should be unlikely.
66 * Transitional: While a real important warning, it fires a lot in current code, so a transition is needed
67 in order to let people fix them before promoting the advice to an important warning.
70 Show/hide a specific warning.
72 Each type of warning can be individually displayed or hidden.
73 The `-w` option takes the name of a warning (displayed at the end of the warning message, between parentheses) to activate it;
74 and "no-{name}" to disable it.
75 It has precedence over -q and -W.
76 Multiple `-w` can be given.
78 To show only `missing-doc` warnings in standard"
80 $ nitc -q -w missing-doc standard
82 To show all warnings and advices, except `missing-doc`:
84 $ nitc -W -w no-missing-doc standard
86 To show important warnings except `useless-type-test`, but not advice except `missing-doc`:
88 $ nitc -w missing-doc -w no-useless-type-test standard
92 May be overridden by `-w`
94 ### `--stop-on-first-error`
95 Just display the first encountered error then stop.
97 By default, nitc tries to detect and display more than one error before aborting the compilation.
100 Do not use color to display errors and warnings.
102 Also, do not echo the line.
103 This options is mainly used by scripts and tools that need parsable error messages.
105 ### `-v`, `--verbose`
106 Additional messages from the tool.
107 Multiple `-v` can be given to improve the verbosity.
109 With one `-v`, there is constant number of lines.
110 With two `-v`, the number of lines is proportional to the number of modules.
111 With three `-v`, the number of lines is proportional to the number of definition of classes.
112 With four `-v`, the number of lines is proportional to the number of definition of properties.
115 Generate various log files.
117 The tool will generate some files in the logging directory (see `--log-dir`).
118 These files are intended to the advanced user and the developers of the tools.
121 Directory where to generate log files.
123 By default the directory is called `logs` in the working directory.
126 ### `-h`, `-?`, `--help`
127 Show Help (the list of options).
130 Show version and exit.
136 Add an additional include path.
138 This option is used to indicate an additional path of a directory containing Nit libraries.
140 The path added with `-I` are searched before those added by the environment variable `NIT_PATH`.
142 May be used more than once.
145 Filename of the generated executable.
147 Indicates the path and name of the produced executable.
149 Note: it is better to use `--dir` if only the directory is important.
150 This way, the platform extension will be correctly set.
152 `-o` is not usable if multiple programs are compiled at once.
157 Produce the executables in the given directory instead of the current directory.
160 Base directory of the Nit installation.
162 Has precedence over the environment variable `NIT_DIR`.
167 Directory used to generate temporary files.
169 By default, it is named `nit_compile` and created in the current directory and destroyed after the compilation.
171 If the option `--compile_dir` or `--no-cc` is used, then the directory is not destroyed and let as is.
174 Do not invoke the C compiler.
176 Files in the compilation directory are generated but the C compiler is not invoked.
178 This option is mainly used to produce C files distributable then compilable on system that do not have a Nit compiler (e.g. embedded system).
179 In this case, it is suggested to also use the options `--dir`, `--compile-dir` and `--semi-global`.
181 $ nitc examples/hello_world.nit --no-cc --dir hello --compile-dir hello --semi-global
183 Will produce a `hello` directory that contains the required C files to finish the compilation.
184 Only the C files required for the program are generated.
185 The final binary will be generated in the same directory.
187 Note that, to be useful, the compilation directory is not destroyed when `--no-cc` is used.
190 Additional module to mix-in.
192 Additional modules are imported and refine the main module of the program.
193 This has basically the same effect than implementing a specific module that imports the main module of the program then each one of the mix-in modules.
194 May be used more than once.
196 This is option is used to weave additional behaviors to existing programs.
197 Modules designated to bring features to programs by refining basic or specialized services, without any intervention of the main program, are good candidates to be used with the `-m` option.
200 An other usage of the `-m` option is to compile program to a specific platform. E.g. `emscripten` or `android`.
202 A last usage is to develop programs as product lines with a main basic module (vanilla) and specific distinct features as flavor modules, then to combine them at compile-time.
204 $ nitc prog_vanilla.nit -m feature_chocolate.nit -m feature_cherry.nit
207 Define a specific property.
209 The `-D` option allows to refine a top-level method at compile-time.
210 This has basically the same effect than implementing a specific module that imports the main module of the program and refines the designated methods.
212 The designated method must be top-level function with no parameters that returns a Bool, an Int or a String.
214 The argument of the `-D` option is "{name}={value}".
215 For Bool, the argument can also be just "{name}", in this case, the value is considered to be `true`.
217 $ nitc foo.nit -D prefix=/opt/foo -D port=8080 -D with_ssl
220 Compile in release mode and finalize application.
222 Currently, this only affect the android platform.
225 Compile in debug mode.
227 Currently removes gcc optimizations.
228 Also preserves the source-files directory for C-debuggers.
230 For more debugging-related options, see also `--hardening` and `NIT_GC_OPTION`
234 ### `nitc` includes distinct compilation modes.
237 Use separate compilation (default mode).
239 In separate compilation, modules are compiled independently of their programs.
240 This makes the recompilation of programs faster since only the modified files need to be recompiled.
243 Use global compilation.
245 The produced executables may become huge and the compilation time is prohibitive.
246 But sometime, they are faster.
248 In practice, `--semi-global` produces nearly as fast but smaller executables.
253 Like `--separate` but use an erasure dynamic typing policy for generics and virtual types.
254 Usually you do not need this, even if you understand the previous sentence.
257 ## SEMI-GLOBAL OPTIMIZATIONS
259 In `--separate` and in `--erasure` modes, some optimization can be gained by relaxing the constraint about
260 the independence on programs.
262 Therefore, with these options, the produced executables may be faster and smaller but the recompilation time
266 Enable all semi-global optimizations.
269 Activate RTA (Rapid Type Analysis).
271 This option only make sense in `--erasure` to enable some semi-global optimizations.
273 RTA is implicitly enabled in `--separate` and `--global`.
275 ### `--inline-coloring-numbers`
276 Inline colors and ids (semi-global).
278 ### `--inline-some-methods`
279 Allow the separate compiler to inline some methods (semi-global).
282 ### `--direct-call-monomorph`
283 Allow the separate compiler to direct call monomorphic sites (semi-global).
286 ### `--direct-call-monomorph0`
287 Allow the separate compiler to direct call monomorphic sites (semi-global).
290 The difference with the non-zero option is internal:
291 with this option, the monomorphism is looked-at on the mmethod level and not at the callsite level.
293 ### `--skip-dead-methods`
294 Do not compile dead methods (semi-global).
297 ## LINK-BOOST OPTIMIZATIONS
299 In `--separate` and in `--erasure` modes, some optimization can be gained by hijacking the linker process.
301 Warning: these optimisations are not portable since they use extra features of the GNU linker `ld`.
302 However, there is very few reasons to not use them if GNU `ld` is available.
305 Enable all link-boost optimizations.
307 ### `--colors-are-symbols`
308 Store colors as symbols instead of static data.
310 By default, the various identifiers used to implement OO-mechanisms are stored as genuine constant static variables.
312 This option uses linker symbols to encode these identifiers.
313 This makes the compiled program faster since less indirections are required to get the values.
314 It also produces executables that are a little bit smaller since static memory does not have to store the colors.
316 ### `--substitute-monomorph`
317 Replace monomorphic trampolines with direct calls.
319 Late-binding is implemented with *trampolines*, that are small functions that just select and jump the to right implementations.
320 If, at link-time, is it known that the target will always by the same implementation then all calls to the trampoline are replaced by
321 direct calls to this single implementation.
323 Note that using trampolines as indirection slows down the executable.
324 However, it is expected that the gain of monomorphic direct-calls overcompensates the additional indirections in polymorphic trampoline-calls.
326 Note: automatically enable option `--trampoline-call`.
328 ## POTENTIAL OPTIMIZATIONS
330 These optimisation are not enabled by default as they are counter-effective in most cases.
333 Guard VFT calls with a direct call.
336 Build a poset of types to create more condensed tables.
338 The drawback is that more time and memory are used by the compilation process.
341 ## DANGEROUS OPTIMIZATIONS
343 The following optimizations disable runtime checks.
344 It means that correct (non-buggy) programs may be slightly faster.
345 It also means that incorrect (buggy) programs may have unspecified behaviors
346 (e.g. formatting your hard drive or killing your cat).
348 In fact, these options are mainly used to bench the compilation results.
351 Disable all tests (dangerous).
353 ### `--no-check-covariance`
354 Disable type tests of covariant parameters (dangerous).
356 ### `--no-check-attr-isset`
357 Disable isset tests before each attribute access (dangerous).
359 ### `--no-check-assert`
360 Disable the evaluation of explicit `assert` and `as` (dangerous).
362 ### `--no-check-autocast`
363 Disable implicit casts on unsafe expression usage (dangerous).
365 ### `--no-check-null`
366 Disable tests of null receiver (dangerous).
368 ### `--no-check-erasure-cast`
369 Disable implicit casts on unsafe return with erasure-typing policy (dangerous).
374 These options are used to debug or to bench the compilation results.
375 Usually you do not need them since they make the generated code slower.
377 ### `--no-shortcut-range`
378 Always instantiate a range and its iterator on 'for' loops.
380 ### `--no-union-attribute`
381 Put primitive attributes in a box instead of an union.
383 ### `--no-shortcut-equal`
384 Always call == in a polymorphic way.
386 ### `--no-tag-primitives`
387 Use only boxes for primitive types.
389 The separate compiler uses tagged values to encode common primitive types like Int, Bool and Char.
390 This option disables tags and forces such primitive values to be boxed.
391 The drawback is that each boxing costs a memory allocation thus increases the amount of work for the garbage collector.
393 However, in some cases, it is possible that this option improves performance since the absence of tags simplify the implementation
394 of OO mechanisms like method calls or equality tests.
396 ### `--no-inline-intern`
397 Do not inline call to intern methods.
399 ### `--colo-dead-methods`
400 Force colorization of dead methods.
402 ### `--no-gcc-directive`
403 Disable advanced gcc directives for optimization.
405 ### `--trampoline-call`
406 Use an indirection when calling.
408 Just add the trampolines of `--substitute-monomorph` without doing any additional optimizations.
413 ### `--no-stacktrace`
414 Disable the generation of stack traces.
416 With this option, the compiled program will not display stack traces on runtime errors.
418 Because stack traces rely on libunwind, this option might be useful in order to generate more portable binaries
419 since libunwind might be non available on the runtime system (or available with an ABI incompatible version).
421 The generated C is API-portable and can be reused, distributed and compiled on any supported system.
422 If the option `--no-stacktrace` is not used but the development files of the library `libunwind` are not available, then a warning will be displayed
423 and stack trace will be disabled.
425 Note that the `--no-stacktrace` option (or this absence) can be toggled manually in the generated Makefile (search `NO_STACKTRACE` in the Makefile).
426 Moreover, the environment variable `NIT_NO_STACK` (see bellow) can also be used at runtime to disable stack traces.
429 Enable dynamic measure of memory usage.
431 Compiled programs will generate a large `memory.log` file that logs all memory allocations.
432 This logs file can then be analyzed with the tool `memplot` from contrib.
435 Generate contracts in the C code against bugs in the compiler.
440 These options can be used to control the fine behavior of the tool.
441 They are useless for a normal user.
443 ### `--disable-phase`
444 Disable a specific phase; use `list` to get the list.
447 Only proceed to parse files.
449 ### `--only-metamodel`
450 Stop after meta-model processing.
452 ### `--ignore-visibility`
453 Do not check, and produce errors, on visibility issues.
456 Do not generate main entry point.
459 Maximum number of lines in generated C files. Use 0 for unlimited.
461 ### `--group-c-files`
462 Group all generated code in the same series of files.
465 Additional options to the `make` command.
467 $ nitc foo.nit --make-flags 'CC=clang' --make-flags 'CFLAGS="-O0 -g"'
469 ### `--typing-test-metrics`
470 Enable static and dynamic count of all type tests.
472 ### `--invocation-metrics`
473 Enable static and dynamic count of all method invocations.
475 ### `--isset-checks-metrics`
476 Enable static and dynamic count of isset checks before attributes access.
478 ### `--tables-metrics`
479 Enable static size measuring of tables used for vft, typing and resolution.
481 ### `--set-dummy-tool`
482 Set toolname and version to DUMMY. Useful for testing.
484 ### `--bash-completion`
485 Generate bash_completion file for this program.
488 Generate a stub manpage in pandoc markdown format.
491 Continue after errors, whatever the consequences.
493 The tool does not stop after some errors but continue until it produces incorrect result, crashes, erases the hard drive, or just continue forever in an infinite loop.
494 This option is used to test the robustness of the tools by allowing phases to progress on incorrect data.
497 Force lazy semantic analysis of the source-code.
499 Analysis of methods is thus done only when required.
500 This option breaks the behavior of most of the tools since errors in methods are undetected until the method is required in some processing.
502 # ENVIRONMENT VARIABLES
505 Base directory of the Nit installation.
507 When the `NIT_DIR` environment variable is set then it specifies the path of the Nit install directory.
509 This directory is used to locate binaries, shared files and the common libraries.
511 When unset, the directory is guessed according to some heuristic.
513 The `--nit-dir` option also set the base directory of the Nit installation but has precedence.
516 Additional include paths.
518 The `NIT_PATH` environment variable contains paths of directories containing Nit libraries.
519 Each path is separated with a column (`:`).
521 The `-I` option also add additional paths.
524 Runtime control of the garbage collector.
526 The behavior of the GC of the executables produced by nitc can be tuned with this environment variable.
528 The environment variable is used when programs are executed, not when they are compiled.
529 Thus, you do not need to recompile programs in order to tweak their GC options.
531 Available values are:
533 * boehm: use the Boehm-Demers-Weiser's conservative garbage collector (default).
534 * malloc: disable the GC and just use `malloc` without doing any `free`.
535 * large: disable the GC and just allocate a large memory area to use for all instantiation.
536 * help: show the list of available options.
539 Runtime control of stack traces.
541 By default, stack traces are printed when a runtime errors occurs during the execution of a compiled program.
542 When setting this environment variable to a non empty value, such stack traces are disabled.
544 The environment variable is used when programs are executed, not when they are compiled.
545 Thus, you do not need to recompile programs in order to disable generated stack traces.
547 Note that stack traces require that, during the compilation, development files of the library `libunwind` are available.
548 If they are not available, then programs are compiled without any stack trace support.
550 To completely disable stack traces, see the option `--no-stacktrace`.
554 The Nit language documentation and the source code of its tools and libraries may be downloaded from <http://nitlanguage.org>