Include statements and compiler linking

J

Joe Hacker

I have what is probably a simple question for some but the process
escapes my understanding. I am a newbee c programmer and understand
some of the basics but I usually just blindly follow directions.
Unfortunatly I'm not the type of person that just does things without
wantint to understand why I have to do certain things.

When you make the typical C program that uses IO using the stdio.h
header file. Why is it that the header file doesnt need to be linked
with the -l argument, but when you are including a library from say a
tutorial or such you also have to name the file in your compile
command.

Is it something to do with the Standard Library already being linked
in with the compiler. If that is the case why do you have to include
it in the source file. I guess I could take another stab at it and
guess that it is because the object code is already linked in by
default for the Standard library. I don't think it is as easy as the
linker not knowing the PATH to the header file.

I appreciate any help guys and look forward to be a active member of
the list.

Thanks
 
B

Barry Schwarz

I have what is probably a simple question for some but the process
escapes my understanding. I am a newbee c programmer and understand
some of the basics but I usually just blindly follow directions.
Unfortunatly I'm not the type of person that just does things without
wantint to understand why I have to do certain things.

When you make the typical C program that uses IO using the stdio.h
header file. Why is it that the header file doesnt need to be linked
with the -l argument, but when you are including a library from say a
tutorial or such you also have to name the file in your compile
command.

Header files are never processed during the link phase.
Is it something to do with the Standard Library already being linked
in with the compiler. If that is the case why do you have to include
it in the source file. I guess I could take another stab at it and

The functions that make up the standard library are never included in
the source program (unless you are actually compiling the source for
one of those functions).
guess that it is because the object code is already linked in by
default for the Standard library. I don't think it is as easy as the
linker not knowing the PATH to the header file.

Think about the contents of a header file and think about what types
of files get processed during the link phase.

Header files contain source code. When you #include a header file,
the contents of that file (that collection of source code) is inserted
into your source code at the point of the #include directive. The
compiler then processes that code exactly the same as it processes the
code you wrote.

During the link phase, object modules (possibly in separate files and
possibly in libraries) are joined together to form an executable. No
source code is ever processed during the link phase.

Whoever builds the implementation (the compiler, the linker, the
standard headers, the standard library functions, etc) gets to decide
how the data needed in the various phases is located. Usually, there
are defaults for where to look for header files and where to look to
resolve references to external functions. If the data you need is not
in the default location, then you need to tell the system where the
data is. To that end some, but only some, implementations use -i and
-l command line arguments. There is nothing standard about these
command line arguments. Each implementation is free to develop its
own conventions.

At least one popular implementation requires the -l argument if you
reference functions declared in math.h. That is nothing more or less
than a choice made when that implementation was being developed. Why
they chose to handle the math functions differently than string
functions and I/O functions and allocation functions, etc is not a
language question but rather one about design philosophy, best asked
in a newsgroup where the implementation is topical.
 
J

jacob navia

Joe Hacker a écrit :
I have what is probably a simple question for some but the process
escapes my understanding. I am a newbee c programmer and understand
some of the basics but I usually just blindly follow directions.
Unfortunatly I'm not the type of person that just does things without
wantint to understand why I have to do certain things.

When you make the typical C program that uses IO using the stdio.h
header file. Why is it that the header file doesnt need to be linked
with the -l argument, but when you are including a library from say a
tutorial or such you also have to name the file in your compile
command.

Is it something to do with the Standard Library already being linked
in with the compiler. If that is the case why do you have to include
it in the source file. I guess I could take another stab at it and
guess that it is because the object code is already linked in by
default for the Standard library. I don't think it is as easy as the
linker not knowing the PATH to the header file.

I appreciate any help guys and look forward to be a active member of
the list.

Thanks

The include files contain text that will be included into your program
text as if you would have typed it in. Nothing less, nothing more. Those
files describe the interface of the included functions and data types to
the compiler without actually implementing them. The implementation is
in some object file (binary compiler output) somewhere.

When linking the program, the compilation system will add some binary
files to the link that you haven't specified. For instance it will
automatically add an object file that contains the code that calls the
"main" function, also called the "startup" code.

Many modern compilers include by default a bunch of commonly used
libraries to avoid making the user tell the linker at each link that he
needs the C library, since it can be assumed that if you program in C
you will use the C library.

Some old compilers and some unfriendly new ones force you to specify the
math library each time you link, provoking endless problems with
newcomers to the language that suddenly realize that sin/cos/sqrt, and
other functions are mising and that you have to specify it with the
cryptic -lm linker command.

Why is that not included? Because for hysterical reasons, many many
years ago, there was a time when computers were so slow that not
including the math library in the link speeded up the linking process.
Today, this doesn't take more than a microsecond more, but hysterical
reasons are just that: HYSTERICAL.

Some compilers like lcc-win or MSVC under windows include a lot of
libraries automatically in the link. They suppose that if you are
programming in windows, libraries like kernel32.lib will be used,
together with other libraries like the one for the common controls, etc.

Some compilers have a directive like

#includelib "foo.lib"

that tells the compiler to add "foo.lib" to the linking process so that
the user doesn't need to remember adding that library to the linker
command line.

In any case, obviously the compilation system can't know about ALL
possible libraries that COULD be used by your program since there are an
infinite number of possible libraries that could exist. At SOME point
you will need to understand that:

(1)
#included files are just text that describes the interface of a module
(2)
Libraries that contain the implementation of a module must be specified
to the linker.

jacob
 
K

Kenny McCormack

jacob navia said:
The include files contain text that will be included into your program
text as if you would have typed it in. Nothing less, nothing more. Those

This the point where a reg will chime in and say words to the effect of:

The so-called "include files" don't need to be files, nor do
they need to be "include"d.

--
No, I haven't, that's why I'm asking questions. If you won't help me,
why don't you just go find your lost manhood elsewhere.

CLC in a nutshell.
 
T

Tom St Denis

This the point where a reg will chime in and say words to the effect of:

        The so-called "include files" don't need to be files, nor do
        they need to be "include"d.

I'd like to point out the irony in that you're both a "reg" (ular pest
that is) and you did indeed post that, thus fulfilling your own
prophecy.

Congrats, you're trolling yourself.

Tom
 
M

Malcolm McLean

This the point where a reg will chime in and say words to the effect of:

        The so-called "include files" don't need to be files, nor do
        they need to be "include"d.
However not infrequently a newbie will say to himself "so include
files are C source included by the compiler in a initial
wordprocessing pass. Now i don't really understand all theis library
malarky. maybe the thing to do is to examine some of the includes to
see if I can figure out what is going on".

It's not inherently a bad stratgey, but the fact is that most include
files for standard or third party libraries are unreadable, if not
absent.
 
B

bart.c

Joe Hacker said:
I have what is probably a simple question for some but the process
escapes my understanding. I am a newbee c programmer and understand
some of the basics but I usually just blindly follow directions.
Unfortunatly I'm not the type of person that just does things without
wantint to understand why I have to do certain things.

When you make the typical C program that uses IO using the stdio.h
header file. Why is it that the header file doesnt need to be linked
with the -l argument, but when you are including a library from say a
tutorial or such you also have to name the file in your compile
command.

A 'header file' is slightly misleading. A header file in C is just an
include file: a source file that is included in another, and typically (for
system headers at least) just contains declarations (not actual code or data
that requires linking).

So an include file does not normally result in an independent object file,
and linking such an include file is meaningless.
Is it something to do with the Standard Library already being linked
in with the compiler. If that is the case why do you have to include
it in the source file.

As I understand how it works, this system seems a mess.

There is a standard library file that needs to be linked in (either
statically or dynamically), but the declarations needed for this are spread
over a couple of dozen 'header files'. In my experience, no special link
options are needed to add in this extra library.

It's possible this standard library may be split into several files, but
there is little correlation with system header files.
I guess I could take another stab at it and
guess that it is because the object code is already linked in by
default for the Standard library. I don't think it is as easy as the
linker not knowing the PATH to the header file.

It's often a mystery to me too. Once a compiler setup works, it's best to
leave well alone.
 
K

Kenny McCormack

I'd like to point out the irony in that you're both a "reg" (ular pest
that is) and you did indeed post that, thus fulfilling your own
prophecy.

You need to learn how to read for comprehension, boy.

If Richard Dawkins, in the middle of some discussion, said something
like "This is where a religious person might say: ...", would any
sensible person think either that:

1) RD was a religious person
2) RD believed the religious nonsense quoted.

Note that *you* (Mr. Tom SD Loony Tunes) might actually think such a
thing, but again, no sensible person would.

And yet, I could imagine some religious loony tune (akin to the
religious loony tunes found in CLC) claiming just that. You hear about
it from time to time in the news.

--
No, I haven't, that's why I'm asking questions. If you won't help me,
why don't you just go find your lost manhood elsewhere.

CLC in a nutshell.
 
K

Kenny McCormack

However not infrequently a newbie will say to himself "so include
files are C source included by the compiler in a initial
wordprocessing pass. Now i don't really understand all theis library
malarky. maybe the thing to do is to examine some of the includes to
see if I can figure out what is going on".

It's not inherently a bad stratgey, but the fact is that most include
files for standard or third party libraries are unreadable, if not
absent.

Really? That doesn't agree with my experience.

--
No, I haven't, that's why I'm asking questions. If you won't help me,
why don't you just go find your lost manhood elsewhere.

CLC in a nutshell.
 
T

Tom St Denis

If Richard Dawkins, in the middle of some discussion, said something
like "This is where a religious person might say: ...", would any
sensible person think either that:

    1) RD was a religious person
    2) RD believed the religious nonsense quoted.

You said "a reg" ... you post here REGULARLY, that makes you a reg.
And you did in fact say it, even if quoted.

Therefore, I stand by my original observation that you're trolling
yourself.

It'd be more like if RD said "This is where an Atheist might say, all
religious folk are nutters." And then you can say he IS an atheist
and he did just say that. IOW if you weren't a reg then your RD
analogy would work, but you are a regular and you did say it.

Tom
 
E

Eric Sosman

When you make the typical C program that uses IO using the stdio.h
header file. Why is it that the header file doesnt need to be linked
with the -l argument, but when you are including a library from say a
tutorial or such you also have to name the file in your compile
command.

This is a characteristic not of the C language, but of the C
implementation you're using. The rules of the language tell you
how to express what the program should do, and that's "the same"
(with some permitted variations) for all implementations. But the
incantations that you use to tell the implementation to translate
your source files, combine them into a program, find necessary
libraries, and run the result are the business of the implementation,
and implementations differ. Sometimes they differ a lot.

Most C implementations (all that I've run into) will find at
least some of the Standard library routines without being told
explicitly where to look for them. This is done as a convenience
for the programmer: Since a very large fraction of C programs will
use the things declared in <stdio.h> and <stdlib.h> and <string.h>
and so on, the implementation "just knows" where to look for them
and will do so "on its own."

For historical reasons, some implementations treat the <math.h>
functions differently: They're found in a conventional place, but
the place won't be searched unless you ask for it with something
like an "-lm" flag. Others treat the <math.h> functions the same
way they treat the <ctype.h> functions, and will find them without
being tickled.

No implementation I know of will automatically search user-
defined libraries without being told to do so, although I suppose
it's possible. The problem (from the implementation's perspective)
is that there might be several user-defined libraries lying around,
perhaps containing several variant versions of the same functions;
how is the implementation to know which library it should search?
Also, for some implementations the search order is important: If
your program calls libFunc() and libFunc() calls atan2(), it might
be important to search the math library after rather than before
searching the library that contains libFunc(). Again, this is a
problem for the implementation: How is it supposed to figure out
which of the N! ways to search N libraries is the right one? So
implementations usually require you to specify (somehow) which of
your own libraries to use and what order to search them in.
Is it something to do with the Standard Library already being linked
in with the compiler. If that is the case why do you have to include
it in the source file. I guess I could take another stab at it and
guess that it is because the object code is already linked in by
default for the Standard library. I don't think it is as easy as the
linker not knowing the PATH to the header file.

The headers like <stdio.h> and <string.h> -- and for that matter
"mystuff.h" -- don't actually contain the functions printf() and
strtok() and myFunc(), but only descriptions of those functions.
"Listen up, compiler," says <stdio.h>, "I'm going to make some calls
to a function named fgets(), which takes three parameters of these
types and returns a value of that type." With this information the
compiler knows what kind of code to generate when it sees an fgets()
call in your source code, but it still doesn't know what the body of
fgets() looks like. It goes ahead and translates your source code,
accepting on faith that it will eventually find the compiled guts of
an fgets() function, somewhere. You get an object file containing
"external references" to functions (and sometimes data) that will be
found elsewhere at link time.

That's the "classical" style, anyhow. Some compilers do in fact
have built-in knowledge of Standard library functions, especially
short functions that might compile to in-line instructions instead of
to full-strength function calls. A call to sqrt(), for example, might
turn into an FSQRT instruction right in the middle of your code, with
no external reference to a separate sqrt() function and no need to
look for one in a library. But these are departures from the norm
for the sake of optimization; for most functions (both the Standard
library's and your own), the headers merely declare and the function
bodies are obtained from somewhere else.
 
W

William Hughes

I have what is probably a simple question for some but the process
escapes my understanding. I am a newbee c programmer and understand
some of the basics but I usually just blindly follow directions.
Unfortunatly I'm not the type of person that just does things without
wantint to understand why I have to do certain things.

When you make the typical C program that uses IO using the stdio.h
header file. Why is it that the header file doesnt need to be linked
with the -l argument, but when you are including a library from say a
tutorial or such you also have to name the file in your compile
command.

Is it something to do with the Standard Library already being linked
in with the compiler. If that is the case why do you have to include
it in the source file. I guess I could take another stab at it and
guess that it is because the object code is already linked in by
default for the Standard library. I don't think it is as easy as the
linker not knowing the PATH to the header file.

I appreciate any help guys and look forward to be a active member of
the list.

Thanks

[File all of this under "Lies to programmers",
i.e. it is not true, but it is a reasonable approximation]

The implementation works more or less like this

-you provide some c code to do something useful

-there is function (e.g printf), that has been written by
someone else, (they may or may not have used c)
instead of writing this yourself you want to just
use it. You tell the compiler how to use the code
by including a header file

-the compiler will then create code that correctly
calls printf, but will not include the code
for printf

-the linker will create an executable by taking the
code provided by the compiler and adding the code for
printf. You can tell it where to find this code by
adding an -lm option.

Now, usually the implementation will not include the header
files for common functions by default, but will search
for the code to common functions by default. So if you want to
use printf you have to #include stdio.h to tell the compiler how
to call printf, but you do not have to -lm libc.h to tell the linker
where to find the code for printf.

Why the difference? Well, there are many common functions
so if you simply #include the instructions to call any
of them, you might make a very small source file very big.
A simple approach is to scan the file
first to find out which functions
are going to be used, then go back and start again, but
this would take time so it was not done by the first
compilers. On the other hand, by the time the linker
gets to things, you know which functions you want, so the linker
only has to put the stuff you need in.

[Could a compiler be written so as to recognize
many common functions, and automatically include the needed
files? Yes. Is this done? No. There are explicit rules about
what should be done if a header file for a function is not
included, and including the right header file could do something
else. There is nothing to stop the compiler warning you that
you are telling it to do something that is almost certainly not
what you want.]

Now, why can't the linker find every function. Well, there
are an infinite number a possible functions, and the code
could be anywhere, so the linker would have to do a very
big search. Worse, the linker would often find more than
one function with the same name. How would it know which
one to use? So if you want to use e.g. a tutorial function,
you have to tell the linker where to find it.

- William Hughes
 
K

Keith Thompson

Joe Hacker said:
I have what is probably a simple question for some but the process
escapes my understanding. I am a newbee c programmer and understand
some of the basics but I usually just blindly follow directions.
Unfortunatly I'm not the type of person that just does things without
wantint to understand why I have to do certain things.

When you make the typical C program that uses IO using the stdio.h
header file. Why is it that the header file doesnt need to be linked
with the -l argument, but when you are including a library from say a
tutorial or such you also have to name the file in your compile
command.
[...]

Header files are compiled, not linked.

Libraries are usually provided in two separate pieces. The header
file, such as "foo.h", merely defines an interface. #include'ing it
makes the declarations visible, so the compiler can generate calls
to the library code, but it doesn't provide the code that implements
the functionality. That code is provided by a binary file that,
depending on the system, might be called "foo.lib", "foo.dll",
"foo.a", "foo.so", or whatever. The compiler, when it's compiling
*your* code, doesn't need this, but the linker, when it combines your
code with the library code to produce an executable program, does.
(I'm ignoring static vs. dynamic linking.)

Typically some or all of the C standard library, including the code
that implements the functions declared in <stdio.h>, will be used
automatically by the linker. (Headers are never used automatically,
so you always need to use #include explicitly.)

So when you write ``#include <stdio.h>'', that tells the compiler
how to generate a call to printf(), but it doesn't provide the
implementation of printf(). That implementation is provided by
the linker, and since it's part of the C standard library, it's
used automatically.

When you write ``#include foo.h'', the same thing happens --
except that the linker doesn't know to use "foo.lib" or whatever
it's called unless you tell it to.
 
K

Keith Thompson

Kenneth Brody said:
I believe that's only true for the Standard header files. It is entirely
possible that

#include <stdio.h>

will not read a file called "stdio.h" from a file somewhere on the
computer's filesystem and insert it as a stream of text into the current
source code. I will note, however, that I have never personally seen such a
compiler. Then again, I've never seen CHARBITS not set to 8, nor a system
with sizeof(int) equal to 1. That doesn't mean they don't exist.

That's essentially correct. It's covered in more detail in C99 6.10.2.

A directive of the form
#include <...>
searches for and includes a "header", which isn't necessarily be a file.
A directive of the form
#include "..."
searches for and includes a "source file". If the search fails, it
then processes it as if it read
#include <...>

The term "source file" strongly implies that it's an actual file,
but it's not entirely clear what that means. It usually refers
to the same kind of entity that can be opened with fopen(), for
example, but it could be anything. For that matter, the compiler
needn't be implemented in C, and its methods for reading headers
and source files needn't have anything to do with C's stdio.
 
S

spinoza1111

Joe Hacker a écrit :










The include files contain text that will be included into your program
text as if you would have typed it in. Nothing less, nothing more. Those
files describe the interface of the included functions and data types to
the compiler without actually implementing them. The implementation is
in some object file (binary compiler output) somewhere.

One hopes.

Sometimes, the include files contain pure or *reinen* preprocessor
material that doesn't correspond to object code at all. And note that
a civilized poster will point this out to enhance the grandeur of your
response, not to "show you're wrong". That is what a code monkey like
Heathfield would do.

Most generalizations have counter-examples. To the small and
technically educated mind like that of Heathfield, the counter-example
falsifies the generalization in all cases. Whereas truly civilized
people such as you and I, mon cher Monsieur, know that only in logic,
that grand tautology, are generalizations air tight and useless for
that reason.

When linking the program, the compilation system will add some binary
files to the link that you haven't specified. For instance it will
automatically add an object file that contains the code that calls the
"main" function, also called the "startup" code.

Many modern compilers include by default a bunch of commonly used
libraries to avoid making the user tell the linker at each link that he
needs the C library, since it can be assumed that if you program in C
you will use the C library.

Some old compilers and some unfriendly new ones force you to specify the
math library each time you link, provoking endless problems with
newcomers to the language that suddenly realize that sin/cos/sqrt, and
other functions are mising and that you have to specify it with the
cryptic -lm linker command.

Why is that not included? Because for hysterical reasons, many many
years ago, there was a time when computers were so slow that not
including the math library in the link speeded up the linking process.
Today, this  doesn't take more than a microsecond more, but hysterical
reasons are just that: HYSTERICAL.

Ah, humor, excellent, Monsieur. Clearly the springtime in Paris works
its magic as you leave your dark chambers for the air of the
boulevards Haussmann, and are again enchanted by la lumiere, la grande
puissance, les femmes charmant of your fair city. As you continue to
work on code, may you continue to repair to the plain and wandering
air.
Some compilers like lcc-win or MSVC under windows include a lot of
libraries automatically in the link. They suppose that if you are
programming in windows, libraries like kernel32.lib will be used,
together with other libraries like the one for the common controls, etc.

Some compilers have a directive like

#includelib "foo.lib"

that tells the compiler to add "foo.lib" to the linking process so that
the user doesn't need to remember adding that library to the linker
command line.

In any case, obviously the compilation system can't know about ALL
possible libraries that COULD be used by your program since there are an
infinite number of possible libraries that could exist. At SOME point
you will need to understand that:

(1)
#included files are just text that describes the interface of a module
(2)
Libraries that contain the implementation of a module must be specified
to the linker.

The #include files are the theory: the object code is the practice.
 
S

spinoza1111

I believe that's only true for the Standard header files.  It is entirely
possible that

    #include <stdio.h>

will not read a file called "stdio.h" from a file somewhere on the
computer's filesystem and insert it as a stream of text into the current
source code.  I will note, however, that I have never personally seen such a
compiler.  Then again, I've never seen CHARBITS not set to 8, nor a system
with sizeof(int) equal to 1.  That doesn't mean they don't exist.

You guys need an International Bac class in theory of knowledge, for
you seem to me to be regressing to a 17th century equation of
knowledge with ignorance, like the Pirate who speaks most wisely and
gravely, look'ee, of Sea Monsters thar be, burn me buttocks.

"Hacker culture" is infantile regression, as witness the popularity
amongst hackers of Creative Anachronism and re-enactment of ancient
slaughters, and the absence of a scientific spirit.
 
T

Tim Rentsch

This the point where a reg will chime in and say words to the effect of:

The so-called "include files" don't need to be files, nor do
they need to be "include"d.

If the supposed statements are informative or potentially helpful,
and also factually accurate, what is the objection to them?
 

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments. After that, you can post your question and our members will help you out.

Ask a Question

Members online

Forum statistics

Threads
473,990
Messages
2,570,211
Members
46,799
Latest member
Mercury_Dev

Latest Threads

Top