GNU Make based build system with a different approach.
GNU Make is an old dependency tracking program. Over the years, it gained a lot of features that are hard to use in a maintainable way. Such as, built-in implicit rules. It also improved in several areas, but these improvements are still optional to be compatible with older Makefiles.
For example:
- GNU Make 4.0 has option to sync output of multiple parallel jobs (-O);
- It also has option to disable built-in implicit recipes and variables (-r and -R);
- One can now use custom recipe prefix instead of TAB character, making for more explicit Makefiles;
- --trace enables tracing of full targets (with prerequisites that caused the target to update).
Let's break backwards compatibility. Let's squeeze every drop of goodness GNU Make has to offer. Let's make a build system that is easier to use and is more featureful.
The user is supposed to never need to call clean goal. The single entry point is qake, and it should bring the project to the most-recent state without doing anything unnecessary.
- Compact syntax for definition of entire program build;
- Prevents lengthy meaningless rebuilds - prunes it as early as possible. Changing the Git branch doesn't cause the full rebuild anymore!;
- Fully parallel. Gets to up-to-date state as quickly as possible. But nothing is done several times; everything is properly included in dependency graph. Even directory creation (see below);
- Tracks build commands - changes of build system itself are taken into account;
- Terse, but useful, output - we value users' attention and think it's easier to spot anything unusual when the build system doesn't spit at terminal every command it runs. At the same time, errors are colorized to be easy to notice and provide the full used command when user needs it the most;
- Extensible to other types of built entities. Supporting builds of static and dynamic libraries with dependencies between them is possible, as well as adding ways to build some custom artifacts;
- Creates needed directories automatically. Just add
| $(DIRECTORY)to prerequisites. This is something you'll notice when adding support of other built artifacts; - Tracks header dependencies (of course) - when user changes a common header, they don't need to remember to
cleanor do anything unusual. They just typeqake, and whatever needs update gets updated; - Uses GNU Make 4.0 internally and exposes it when needed - you can still pass custom flags to Make itself for debugging or any other purpose.
The project is in prototype stage and may or may not have significant limitations, including those making it unusable for a particular application.
For one, current implementation of object and source files tracking uses full hashing of corresponding file, what decreases the performance significantly.
You need to call PROGRAM and eval the result:
$(eval $(call PROGRAM,\
,\ # name of directory with sources under 'src'
program,\ # name of directory with results under 'build/res'
$(SRC),\ # list of sources
$(CFLAGS),\ # compiler flags for compilation of objects
$(LDFLAGS),\ # compiler flags for linking of objects to program
$(LDLIBS),\ # libraries for proper linking of objects to program
))in your Makefile.
For this to work, you need Qake to be installed (see below). You also need yo define SRC variable to a list of paths to sources, relative to root directory of project. In case all your sources are directly in src, that's going to be src/a.c src/b.c ...:
SRC := $(wildcard src/*.c)And a small touch on top: for tracking of commands used to build your program, we need the name of Makefile defining the build:
THIS_MAKEFILE := $(lastword $(MAKEFILE_LIST))This code is always the same, but you need to put it at the beginning of the Makefile, as MAKEFILE_LIST gets updated by Make as it goes and includes something.
That's it! With this, you get make all and make clean, with ability to separately build objects via path reference:
make build/res/awesome_object.c.oFor concrete example, see tests/circle/Makefile. The rest of the tour will assume interaction with build of that program (it's an IRC chat named circle).
There's also a test.sh, which tests all the supposedly working modes of the build. You can run it from the tests/circle/ directory.
➜ circle git:(master) ✗ qake
MKDIR build/res/circled
MKDIR build/aux
MKDIR build/aux/circled
GCC irclist.c.o
GCC irc.c.o
GCC ircfunc.c.o
GCC ircenv.c.o
GCC ircsock.c.o
GCC main.c.o
GCC ircq.c.o
GCC circled
➜ circle git:(master) ✗As you can see, the default logging is terse and produces only the short description of command with name of a target.
This makes sense to not pollute the screen, and to make warnings easily detectable.
There currently is no "verbose" mode support (so that commands are output in full as in usual Make), but its' addition is trivial.
In case there's a compilation error, the build will stop and print the entire attempted command, so you'll be able to easily copy-paste it and retry it in manual mode:
➜ circle git:(master) ✗ qake
GCC irc.c.o
src/irc.c:536:9: error: redefinition of ‘__irc_get_kicked_nick’
field_t __irc_get_kicked_nick (const char * message)
^
src/irc.c:518:9: note: previous definition of ‘__irc_get_kicked_nick’ was here
field_t __irc_get_kicked_nick (const char * message)
^
src/irc.c: In function ‘__irc_get_kicked_nick’:
src/irc.c:538:5: error: incompatible types when returning type ‘void *’ but ‘field_t’ was expected
return NULL;
^
[ERROR] Failed command: gcc src/irc.c -o build/res/circled/irc.c.o -c -MD -MF build/aux/circled/irc.c.o.d -MP
➜ circle git:(master) ✗I can't show it here, but [ERROR] is in red on any modern terminal, so the errors will really stand out.
And of course, null-build at this point is performed instantly and doesn't rebuild anything:
➜ circle git:(master) ✗ qake
➜ circle git:(master) ✗ time qake
/home/mpankov/qake//qake 0,07s user 0,07s system 112% cpu 0,127 total
➜ circle git:(master) ✗Let's start with basics: tracking of included headers.
You can edit the header and then everything depending on it will get rebuilt:
➜ circle git:(master) ✗ qake
➜ circle git:(master) ✗ emacs src/ircfunc.h
Waiting for Emacs...
➜ circle git:(master) ✗ qake
GCC main.c.o
GCC ircfunc.c.o
GCC irclist.c.o
GCC ircsock.c.o
GCC ircenv.c.o
GCC ircq.c.o
GCC irc.c.o
GCC circled
➜ circle git:(master) ✗Now, to some more magical features. You can change the build command of program, for example, by changing the LDFLAGS:
➜ circle git:(master) ✗ qake
➜ circle git:(master) ✗ emacs Makefile
Waiting for Emacs...in the following way:
LDLIBS_CIRCLE := \
-ldl \
-lpthread \and the build notices that! See:
➜ circle git:(master) ✗ qake
GCC circled
➜ circle git:(master) ✗This is done so that the user never has to think about whether they need to clean anything. You simply use qake in all situations, even if you change the description of the build itself.
More cool magic to come! In case you modify the source:
➜ circle git:(master) ✗ qake
➜ circle git:(master) ✗ emacs src/irc.c
Waiting for Emacs...by adding the following:
// This is a super-useful comment about some function.and re-build:
➜ circle git:(master) ✗ qake
GCC irc.c.o
➜ circle git:(master) ✗then you'll notice only the corresponding object file got rebuilt. What is it, a bug?.. Well, not actually - it's a way to cut propagation of meaningless changes further down the dependency graph.
Surely enough, object file we got after adding the comment is the same as before, so why re-link the program?
In more complex cases, this optimization allows us to produce nearly-null builds in situations where a full rebuild would happen in usual Make build. Just imagine several programs depending on that tiny object. Or possibly something uses the compiled program for other automated build actions, which would have to be re-run, too...
And just to make sure meaningful changes still propagate properly:
➜ circle git:(master) ✗ emacs src/irc.c
Waiting for Emacs...we add a new function:
void awesome_func(void)
{
printf("42");
}and rebuild then:
➜ circle git:(master) ✗ qake
GCC irc.c.o
GCC circled
➜ circle git:(master) ✗Yay! We've rebuilt the program when needed.
It not only saves us from long rebuilds when you, say, change just the documentation. It also saves us from rebuilding when Git branch changes, or somebody touches the file accidentally, etc.
You can install qake via the command line with wget. git is also required.
GNU Make 4.0 will be downloaded, built and installed during to the installation.
If you're using wget type:
wget --no-check-certificate https://github.com/mkpankov/qake/raw/master/installer.sh -O - | sh