MACROs are ugly but...

[Ben Hetland]

....in certain cituations they can be useful if not for anything else,
then at least for saving a lot of repetetetetetitititive typing.

Beyond the point of "do something better instead", I'm curious about how
the following syntactical problem can be solved. It should apply equally
to C and C++ as it mainly is a preprocessor-related problem.

I tryed to define something similar to the following example:

-------------------------------------
#define DO_DIRTYWORK \
\
#pragma warning( disable : 4995 ) /* Annoying warning */ \
\
/* Do some dirty work here (that I want to hide) */ \
\
#pragma warning( default : 4995 ) /* Warning reactivated */ \
/* finished dirty part */
-------------------------------------

The point was of course to use that DO_DIRTYWORK several places around
the sources. Its content was not really part of the main logic, but
merely a library workaround that required some lines of code injected
"everywhere". (When the library was fixed later, then maybe I could just
redefine the DO_DIRTYWORK macro as empty!)

However, this doesn't compile of course, because a prepocessor directive
needs to start as the first non-blank # on a line, which is conflicting
with the macro's end-of-line escapes (the \). Obviously, what I _wanted_
to achieve in this particular case, was to have the actual preprocessor
directives emitted from the preprocessor's first replacement (then
further preprocessed in-place where the actual code is used).

Can something like that be achieved?
How?

 

[David Resnick]

...in certain cituations they can be useful if not for anything else,
then at least for saving a lot of repetetetetetitititive typing.

Beyond the point of "do something better instead", I'm curious about how
the following syntactical problem can be solved. It should apply equally
to C and C++ as it mainly is a preprocessor-related problem.

I tryed to define something similar to the following example:

-------------------------------------
#define DO_DIRTYWORK \
\
#pragma warning( disable : 4995 ) /* Annoying warning */ \
\
/* Do some dirty work here (that I want to hide) */ \
\
#pragma warning( default : 4995 ) /* Warning reactivated */ \
/* finished dirty part */
-------------------------------------

The point was of course to use that DO_DIRTYWORK several places around
the sources. Its content was not really part of the main logic, but
merely a library workaround that required some lines of code injected
"everywhere". (When the library was fixed later, then maybe I could just
redefine the DO_DIRTYWORK macro as empty!)

However, this doesn't compile of course, because a prepocessor directive
needs to start as the first non-blank # on a line, which is conflicting
with the macro's end-of-line escapes (the \). Obviously, what I _wanted_
to achieve in this particular case, was to have the actual preprocessor
directives emitted from the preprocessor's first replacement (then
further preprocessed in-place where the actual code is used).

Can something like that be achieved?
How?

How about putting the "dirtywork" in a separate header, and doing this:

#if defined(DO_DIRTYWORK)
#include "dirtywork.h"
#endif

Then have your pragmas and whatnot in your dirtywork.h...

-David

 

[Bob Hairgrove]

[cross-posting removed]

How about putting the "dirtywork" in a separate header, and doing this:

#if defined(DO_DIRTYWORK)
#include "dirtywork.h"
#endif

Then have your pragmas and whatnot in your dirtywork.h...

-David

This is a pretty good solution for configuration macros which are
defined for an entire project, or for a big part of one.

Unfortunately, it isn't quite as easy as that. Often one would want to
wrap only single functions, or perhaps the contents of a single source
file (not a header!) between #pragma warning (disable ... ) and
#pragma warning ( default ... ) or something similar. This would
require at least two separate header files, and I think the
maintenance involved would be greater than just typing in the
pragma's.

IMHO one should only put such things into separate files which need to
be accessed by more than one source file. One would hardly wish to
turn off specific warnings for an entire project, although there are
certainly enough programmers who would argue otherwise. I, for one, am
only happy when my code compiles without any warnings at all. If there
is a certain warning I cannot work around, then I will turn it off
only for whatever portion of code causes the warning and document the
place well as to why it should be turned off. Typing the #pragma or
whatever directly into the source enforces the idiom that they are
extremely local, and that there is a strong dependency between them
and the code they influence.

At any rate, commonly-used #pragmas can easily be stored as templates
in most modern-day IDE's to save typing.

 

[Greg]

...in certain cituations they can be useful if not for anything else,
then at least for saving a lot of repetetetetetitititive typing.

Beyond the point of "do something better instead", I'm curious about how
the following syntactical problem can be solved. It should apply equally
to C and C++ as it mainly is a preprocessor-related problem.

I tryed to define something similar to the following example:

-------------------------------------
#define DO_DIRTYWORK \
\
#pragma warning( disable : 4995 ) /* Annoying warning */ \
\
/* Do some dirty work here (that I want to hide) */ \
\
#pragma warning( default : 4995 ) /* Warning reactivated */ \
/* finished dirty part */
-------------------------------------

The point was of course to use that DO_DIRTYWORK several places around
the sources. Its content was not really part of the main logic, but
merely a library workaround that required some lines of code injected
"everywhere". (When the library was fixed later, then maybe I could just
redefine the DO_DIRTYWORK macro as empty!)

However, this doesn't compile of course, because a prepocessor directive
needs to start as the first non-blank # on a line, which is conflicting
with the macro's end-of-line escapes (the \). Obviously, what I _wanted_
to achieve in this particular case, was to have the actual preprocessor
directives emitted from the preprocessor's first replacement (then
further preprocessed in-place where the actual code is used).

Can something like that be achieved?
How?

You should be able to use the _Pragma preprocessor operator to create
#pragma directives using macros. _Pragma is part of the C99 standard
and I am unsure of its current status in C++. But many C++ compilers
implement it.

Greg

 

[Ben Hetland]

How about putting the "dirtywork" in a separate header, and doing this:

#if defined(DO_DIRTYWORK)
#include "dirtywork.h"
#endif

Yes, that might work, but means 3 lines of code for every single of
these places that I need this thing. I think that is just contributing
more to "polluting" the general code structure/algorithm wherever it's
used. BTW, if it's only the one pragma, then it could also be achieved
without the extra include by writing

#pragma ( disable ... )
DO_DIRTYWORK
#pragma ( default ... )

everywhere. IMHO, it's still a bit too messy.
(However, the real-life case had more than that...)

I guess I cannot even put it all into an include file, and have the
macro #include that, because of the # again...

 

[mlimber]

...in certain cituations they can be useful if not for anything else,
then at least for saving a lot of repetetetetetitititive typing.

Beyond the point of "do something better instead", I'm curious about how
the following syntactical problem can be solved. It should apply equally
to C and C++ as it mainly is a preprocessor-related problem.

I tryed to define something similar to the following example:

-------------------------------------
#define DO_DIRTYWORK \
\
#pragma warning( disable : 4995 ) /* Annoying warning */ \
\
/* Do some dirty work here (that I want to hide) */ \
\
#pragma warning( default : 4995 ) /* Warning reactivated */ \
/* finished dirty part */
-------------------------------------

The point was of course to use that DO_DIRTYWORK several places around
the sources. Its content was not really part of the main logic, but
merely a library workaround that required some lines of code injected
"everywhere". (When the library was fixed later, then maybe I could just
redefine the DO_DIRTYWORK macro as empty!)

However, this doesn't compile of course, because a prepocessor directive
needs to start as the first non-blank # on a line, which is conflicting
with the macro's end-of-line escapes (the \). Obviously, what I _wanted_
to achieve in this particular case, was to have the actual preprocessor
directives emitted from the preprocessor's first replacement (then
further preprocessed in-place where the actual code is used).

Can something like that be achieved?
How?

A small criticism: if possible, it would be better to save and restore
the state of the warnings rather than set them to the default since the
warnings may have already enabled or disabled explicitly elsewhere.
With VC++ this can be done with something like this:

#pragma warning(push)
#pragma warning(disable:4995)
// Code that generates 4995 goes here
#pragma warning(pop)

Cheers! --M

 

[Eric Sosman]

Ben Hetland wrote On 10/06/05 09:59,:

...in certain cituations they can be useful if not for anything else,
then at least for saving a lot of repetetetetetitititive typing.

Beyond the point of "do something better instead", I'm curious about how
the following syntactical problem can be solved. It should apply equally
to C and C++ as it mainly is a preprocessor-related problem.

#pragma looks like a preprocessor directive, and the C
Standard describes it in the preprocessor section. If you
think about it, though, it's immediately clear that most
#pragmas are not "really" preprocessor constructs, because
their effect is felt long after the preprocessor has been
and gone. A #pragma that adjusts struct padding or controls
the aggressiveness of the optimizer's loop unrolling has
nothing to do with the preprocessor and everything to do
with code generation.

The point of all this is that the way #pragma is handled
may very well differ between C and C++. Since my knowledge
of C++ wouldn't even earn me a D- I'll answer only for C;
someone else may be able to help you with That Other Language.

I tryed to define something similar to the following example:

-------------------------------------
#define DO_DIRTYWORK \
\
#pragma warning( disable : 4995 ) /* Annoying warning */ \
\
/* Do some dirty work here (that I want to hide) */ \
\
#pragma warning( default : 4995 ) /* Warning reactivated */ \
/* finished dirty part */
-------------------------------------

The point was of course to use that DO_DIRTYWORK several places around
the sources. Its content was not really part of the main logic, but
merely a library workaround that required some lines of code injected
"everywhere". (When the library was fixed later, then maybe I could just
redefine the DO_DIRTYWORK macro as empty!)

However, this doesn't compile of course, because a prepocessor directive
needs to start as the first non-blank # on a line, which is conflicting
with the macro's end-of-line escapes (the \). Obviously, what I _wanted_
to achieve in this particular case, was to have the actual preprocessor
directives emitted from the preprocessor's first replacement (then
further preprocessed in-place where the actual code is used).

The real reason this doesn't work is that expanding
a macro never produces a preprocessor directive, not even
if the expansion happens to resemble one. (6.10.3.4/3)

Can something like that be achieved?
How?

If your compiler supports the latest "C99" Standard, you
can use the _Pragma operator (6.10.9):

#define DO_DIRTYWORK \
_Pragma("warning( disable : 4995 )") \
dirty_little_doings_going_on \
_Pragma("warning( default : 4995 )")

Failing that, the best I can suggest is to put the
dirty work in a separate function, and have DO_DIRTYWORK
generate a call to it.

 

[Alf P. Steinbach]

* Ben Hetland:

...in certain cituations they can be useful if not for anything else,
then at least for saving a lot of repetetetetetitititive typing.

Beyond the point of "do something better instead", I'm curious about how
the following syntactical problem can be solved. It should apply equally
to C and C++ as it mainly is a preprocessor-related problem.

Your abstraction of your proposed _solution_ of the problem is only
preprocessor-related, but whether problem is, is another question.

I tryed to define something similar to the following example:

-------------------------------------
#define DO_DIRTYWORK \
\
#pragma warning( disable : 4995 ) /* Annoying warning */ \
\
/* Do some dirty work here (that I want to hide) */ \
\
#pragma warning( default : 4995 ) /* Warning reactivated */ \
/* finished dirty part */
-------------------------------------

Macros can't generate macros.

But say that warning 4995, for this compiler (whichever it is), is about using
a deprecated function: then that is the problem, and you can ignore the
proposed solution above and concentrate on the problem.

E.g., you can put the usage of a deprecated function in its own module (header
plus implementation), by wrapping it in a similar function, and put the
#pragma -- which in general should be enclosed in #ifdef...#endif compiler
version checking -- at the top of the implementation file.

This is a much better solution because it abstracts the use of the deprecated
function.

Which not only can make the client code more clear, but means you only have to
change _one_ place when that deprecated function is actually removed.

 

[Walter Roberson]

David Resnick wrote:

Yes, that might work, but means 3 lines of code for every single of
these places that I need this thing.

Just

#include "dirtywork.h"

and it will have within it the DO_DIRTYWORK conditional test
and the dirty code.

 

[archivaty]

Ben Hetlandwrote
The point was of course to use that DO_DIRTYWORK several place
aroun

the sources.

Why not having dirtywork in a separated function and then call i
whenever you need to

Sent via Archivaty.com

 

[Ben Hetland]

Your abstraction of your proposed _solution_ of the problem is only
preprocessor-related, but whether problem is, is another question.

That is true, and for instance for C++ the problem could also be solved
using templates in combination with macros in a single header-file.
However, I posted the question to challenge my shortcoming with the
preprocessor's syntax rules, of which some useful comments have been
posted I think. I would also like to have something that could work in
both languages.

Macros can't generate macros.

This is not about generating macros, but about using other preprocessor
features within a preprocessor feature itself, like a macro definition is.

However, I've seen a similar thing that comes very close to generating
macros many times. That is when macros are built from other macros, and
the building of those macros relies on conditional compilations (another
preprocessor feature) to decide how the macro is actually composed. And
that could happen in multiple levels of "macro nesting". So at least the
preprocessor can generate macros...

But say that warning 4995, for this compiler (whichever it is), is about using
a deprecated function: then that is the problem, and you can ignore the
proposed solution above and concentrate on the problem.

Yes, the original problem was with one of Microsoft's compilers, as you
probably guessed. And deprecated functions are already isolated to a few
files. I used this only for illustration. The specific problem also
included other pragmas and other preprocessor features like #ifdef, but
I thought that wasn't so important in illustrating my question.

This is a much better solution because it abstracts the use of the deprecated
function.

I definitely agree with much of this, but just would like to comment
that not only functions may be deprecated, but also types for instance
(as can classes in C++). Types often have a tendency to pollute the rest
of the "client code" with its presense far beyond that isolated module.

Also, as happens with functions as well, prototypes usually need to be
included in header files which are again included in the "client code".
Fine, except when a prototype uses a deprecated type, which is when
"pollution" easily occurs.

I guess it is often due to some laziness among us that things like that
happen, because with careful considerations in the first place it must
be possible to avoid that kind of practice. However, it is just a
real-life observation, and it really requires great skill and foresight
to be able to foresee everything that will become deprecated in the
future. Say, which one of us can claim that they will only need to
modify one of their code units completely independently of all other
source code in case of the hypothetical event that (say) 'size_t' would
become deprecated one day?

 

[Alf P. Steinbach]

* Ben Hetland:

That is true, and for instance for C++ the problem could also be solved
using templates in combination with macros in a single header-file.
However, I posted the question to challenge my shortcoming with the
preprocessor's syntax rules, of which some useful comments have been
posted I think. I would also like to have something that could work in
both languages.

Cross-posted to clc and clc++, I didn't notice.

Cross-language solutions for C++ and C just mean C.

So I think the original question doesn't belong in clc++, but it's a bit late
to fix that now.

This is not about generating macros, but about using other preprocessor
features within a preprocessor feature itself, like a macro definition is.

Yes. Actually what I wrote, as opposed to what I meant, is blatantly false...

However, I've seen a similar thing that comes very close to generating
macros many times. That is when macros are built from other macros, and
the building of those macros relies on conditional compilations (another
preprocessor feature) to decide how the macro is actually composed. And
that could happen in multiple levels of "macro nesting". So at least the
preprocessor can generate macros...

For C++, see the amazing "make the preprocessor jump through hoops and give a
rock n' roll concert" library -- or whatever it should be called -- at

...
Also, as happens with functions as well, prototypes usually need to be
included in header files which are again included in the "client code".
Fine, except when a prototype uses a deprecated type, which is when
"pollution" easily occurs.

For C++, one known solution to that is the PIMPL a.k.a. "compiler firewall"
a.k.a many other things idiom.
<url: http://www.gotw.ca/gotw/024.htm>.
<url: http://www.gotw.ca/gotw/028.htm>.
<url: http://www.gotw.ca/gotw/059.htm> (where the use of std::auto_ptr for an
incomplete type is incorrect; this is the most infamous example of that, the
one often cited when the topicc pops up, still not fixed!).