After all that theory and talking, let's start by building the code written through the last nine parts of this series. This part of our series might actually serve you even if you learned C someplace else, or if you think your practical side of C development needs a little strength. We will see how to install necessary software, what said software does and, most important, how to transform your code into zeros and ones. Before we begin, you might want to take a look at our most recent articles about how to customize your development environment:
Remember the first part of our C Development series? There we outlined the basic process that takes place when you compile your program. But unless you work in compiler development or some other really low level stuff, you won't be interested how many JMP instructions the generated assembler file has, if any. You will only want to know how to be as efficient as possible. This is what this part of the article is all about, but we are only scratching the surface, because of the extensiveness of the subject. But an entry-level C programmer will know after reading this everything needed to work efficiently.
Besides knowing exactly what you want to achieve, you need to be familiar with the tools to achieve what you want. And there is a lot more to Linux development tools than gcc, although it alone would be enough to compile programs, but it would be a tedious task as the size of your project increases. This is why other instruments have been created, and we'll see here what they are and how to get them. I already more than suggested you read the gcc manual, so I will only presume that you did.
Imagine you have a multi-file project, with lots of source files, the works. Now imagine that you have to modify one file (something minor) and add some code to another source file. It would be painful to rebuild all the project because of that. Here's why make was created: based on file timestamps, it detects which files need to be rebuilt in order to get to the desired results (executables, object files...), named targets. If the concept still looks murky, don't worry: after explaining a makefile and the general concepts, it will all seem easier, although advanced make concepts can be headache-inducing.
make has this exact name on all platforms I worked on, that being quite a lot of Linux distros, *BSD and Solaris. So regardless of what package manager you're using (if any), be it apt*, yum, zypper, pacman or emerge, just use the respective install command and make as an argument and that's it. Another approach would be, on distros with package managers that have group support, to install the whole C/C++ development group/pattern. Speaking of languages, I wanted to debunk a myth here, that says makefiles (the set of rules that make has to follow to reach the target) is only used by C/C++ developers. Wrong. Any language with a compiler/interpreter able to be invoked from the shell can use make's facilities. In fact, any project that needs dependency-based updating can use make. So an updated definition of a makefile would be a file that describes the relationships and dependencies between the files of a project, with the purpose of defining what should be updated/recompiled in case one or more files in the dependency chain changes. Understanding how make works is essential for any C developer who works under Linux or Unix - yes, commercial Unix offers make as well, although probably some version that differs from GNU make, which is our subject. "Different version" means more than numbers, it means a BSD makefile is incompatible with a GNU makefile. So make sure you have GNU make installed if you're not on a Linux box.
In the first part of this article, and some subsequent ones, we used and talked about parts of yest, a small program that displays yesterday's date by default, but does a lot of nifty date/time-related things. After working with the author, Kimball Hawkins, a small makefile was born, which is what we'll be working with.
First, let's see some basics about the makefile. The canonical name should be GNUmakefile, but if no such file exists it looks for names like makefile and Makefile, in that order, or so the manual page says. By the way, of course you should read it, and read it again, then read it some more. It's not as big as gcc's and you can learn a lot of useful tricks that will be useful later. The most used name in practice, though, is Makefile, and I have never seen any source with a file named GNUmakefile, truth be told. If, for various reasons, you need to specify another name, use make's -f, like this:
$ make -f mymakefile
Here's yest's Makefile, that you can use to compile and install said program, because it's not uploaded of Sourceforge yet. Although it's only two-file program - the source and the manpage - you will see make becomes useful already.
# Makefile for compiling and installing yest UNAME := $(shell uname -s) CC = gcc CFLAGS = -Wall CP = cp RM = rm RMFLAGS = -f GZIP = gzip VERSION = yest-184.108.40.206 yest: ifeq ($(UNAME), SunOS) $(CC) -DSUNOS $(CFLAGS) -o yest $(VERSION).c else $(CC) $(CFLAGS) -o yest $(VERSION).c endif all: yest install maninstall install: maninstall $(CP) yest /usr/local/bin maninstall: $(CP) $(VERSION).man1 yest.1 $(GZIP) yest.1 $(CP) yest.1.gz /usr/share/man/man1/ clean: $(RM) $(RMFLAGS) yest yest.1.gz deinstall: $(RM) $(RMFLAGS) /usr/local/bin/yest /usr/share/man/man1/yest1.gz
If you look carefully at the code above, you will already observe and learn a number of things. Comments begin with hashes, and since makefiles can become quite cryptic, you better comment your makefiles. Second, you can declare your own variables, and then you can make good use of them. Next comes the essential part: targets. Those words that are followed by a colon are called targets, and one use them like
make [-f makefile name] target_name. If you ever installed from source, you probably typed 'make install'. Well, 'install' is one of the targets in the makefile, and other commonly-used targets include 'clean', 'deinstall' or 'all'. Another most important thing is that the first target is always executed by default if no target is specified. In our case, if I typed 'make', that would have been the equivalent of 'make yest', as you can see, which means conditional compilation (if we are on Solaris/SunOS we need an extra gcc flag) and creation of an executable named 'yest'. Targets like 'all' in our example are doing nothing by themselves, just tell make that they depend on other files/targets to be up to date. Watch the syntax, namely stuff like spaces and tabs, as make is pretty pretentious about things like this.
Here's a short makefile for a project that has two source files. The filenames are src1.c and src2.c and the executable's name needs to be exec. Simple, right?
exec: src1.o src2.o gcc -o exec src1.o src2.o src1.o: src1.c gcc -c src1.c src2.o: src2.c gcc -c src2.c
The only target practically used, which is also the default, is 'exec'. It depends on src1.o and src2.o, which, in turn, depend on the respective .c files. So if you modify, say, src2.c, all you have to do is run make again, which will notice that src2.c is newer than the rest and proceed accordingly. There is much more to make than covered here, but there is no more space. As always, some self-study is encouraged, but if you only need basic functionality, the above will serve you well.
Usually it's not just 'make && make install', because before those two there exists a step that generates the makefile, especially useful when dealing with bigger projects. Basically, said script checks that you have the components needed for compilation installed, but also takes various arguments that help you change the destination of the installed files, and various other options (e.g. Qt4 or GTK3 support, PDF or CBR file support, and so on). Let's see in a short glance what those configure scripts are all about.
You don't usually write the configure script by hand. You use autoconf and automake for this. As the names imply, what they do is generate configure scripts and Makefiles, respectively. For example, in our previous example with the yest program, we actually could use a configure script that detects the OS environment and changes some make variables, and after all that generates a makefile. We've seen that the yest makefile checks if we're running on SunOS, and if we are, adds a compiler flag. I would expand that to check if we're working on a BSD system and if so, invoke gmake (GNU make) instead of the native make which is, as we said, incompatible with GNU makefiles. Both these things are done by using autoconf: we write a small
configure.in file in which we tell autoconf what we need to check, and usually you will want to check for more than OS platform. Maybe the user has no compiler installed, no make, no development libraries that are compile-time important and so on. For example, a line that would check the existence of time.h in the system standard header locations would look like so:
We recommend you start with a not-too-big application, check the source tarball contents and read the configure.in and/or configure.ac files. For tarballs that have them, Makefile.am is also a good way to see how an automake file looks. There are a few good books on the matter, and one of them is Robert Mecklenburg's "Managing Projects with GNU Make".
I know the gcc manual is big and I know many of you haven't even read it. I take pride in reading it all (all that pertains to IA hardware anyway) and i must confess I got a headache afterwards. Then again, there are some options you should know, even though you will learn more as you go.
You have already encountered the -o flag, that tells gcc what the resulting outfile, and -c, that tells gcc not to run the linker, thus producing what the assembler spits out, namely object files. Speaking of which, there are options that control the stages at which gcc should stop execution. So to stop before the assembly stage, after the compilation per se, use -S. In the same vein, -E is to be used if you want to stop gcc right after preprocessing.
It's a good practice to follow a standard, if not for uniformity, but for good programming habits. If you're in the formative period as a C developer, choose a standard (see below) and follow it. The C language was standardized first after Kernighan and Ritchie (RIP) published "The C Programming Language" in 1978. It was a non-formal standard, but in was shortly dubbed K&R and respected. But now it's obsolete and not recommended. Later, in the '80s and the '90s, ANSI and ISO developed an official standard, C89, followed by C99 and C11. gcc also supports other standards, like gnuxx, where xx can be 89 or 99, as examples. Check the manual for details, and the option is '-std=', "enforced" by '-pedantic'.
Warnings-related options start with "-W", like '-Wall' (it tells gcc to enable all errors, although they're not quite all enabled) or '-Werror' (treat warnings as errors, always recommended). You can pass supplemental arguments to the programs that help with the intermediary steps, such as preprocessor, assembler or linker. For example, here's how to pass an option to the linker:
$ gcc [other options...] -Wl,option [yet another set of options...]
Similarly and intuitively, you can use 'Wa,' for the assembler and 'Wp,' for the preprocessor. Take note of the comma and the white space that tells the compiler that the preprocessor/assembler/linker part has ended. Other useful families of options include '-g' and friends for debugging, '-O' and friends for optimization or '-Idirectory' - no white space - to add a header-containing location.
I recommend you take your time to read this article, play with the examples, then write your own, increasing the complexity as you go.
Here is what you can expect next: