Indeed there's only one object file with header-guards.
It's not intended to be used as a library or something.
Moreover - the codebase is really tiny.
That being said - it made with simplicity in mind
Sure, but... wouldn't it have been simpler to have the functions in the .c file?
This is just a odd way of doing things. There are good reasons you don't usually see this, add one more .c file and things get awkward real fast.
Sure, it works in this particular instance, but it just makes me feel itchy. Maybe it's past trauma where I saw real weird codebases with static functions in .h files used across multiple .c files. It was just horrible to work with and reason about.
Well if there are already two remarks on that matter - I'll fix the issue! I know the profits, yeah. Things haven't exactly worked out the way I thought they would - there should be a border between simplicity/bad design.
Hahah, yeah it smells, can't argue with that. I believe that everyone had a trauma like that in the past :)
Haha. Yup, I know that, of course. And I have never used that practice in real world projects. Well, I'll put both .c/.h files, that doesn't add that much overhead.
A is something I want to use in multiple files, like a method declaration, a type, a const or a preprocessor define.
B is a place that I want to use it, and it's tightly related to A.
Unfortunately, B belongs in a .c file.
But to use A in multiple places, it belongs in a .h file.
Crap, now these two very related things that I wanted to keep together lexically must now be split into separate files.
In this case, the author is making a super-simple example project. I'm happy they chose to forgo the standard C header/source file dance. It made it a lot easier to read and understand.
But yeah, the dance is required for "real" projects, as you've described.
- If you need a function visible outside, put it's _prototype_ in the .h file.
- If you want to have a variable visible outside, also make a extern declaration inside the .h file.
- If you want to have anything else visible to the outside, just move it to the .h file.
The reason I say that putting functions inside a .h file is a pain is because of the way the preprocessor works. It takes the contents of the included file and pastes it at the place of the include. So doing a function inside a .h file and using it inside multiple .c files will mean you have the same function multiple times and it will throw a linker error if the function is not static. At link time the linker will be confused because it will not know about witch of the functions you're talking about when you call it. So then you declare it static, but then you _actually_ have 2x functions inside your executable, that do the same thing.
This is even worse with variables declared static inside .h files since here you will think you're changing your variable, but you actually have separate variables inside each compiled file.
Of course, my observation is also based on taste. I like when I have everything inside one file, instead of spread over multiple files( of course, as long as this is reasonable). This program here might have had one single .c file and it would have been just file. I actually would have considered it easier to read, but as I said, it's a matter of taste at this point.
Duplicating some code in the executable of this size might not be that much of a deal. I wonder if there are examples where such practice resulted in improved performance due to locality of the code. Of course you could also toy with inlines.
>So doing a function inside a .h file and using it inside multiple .c files will mean you have the same function multiple times and it will throw a linker error if the function is not static.
This is prevented by a header guard.
And, also: if you have the same function defined elsewhere, and your compiler complains about it: good. That is a bug.
Single-header projects like this are useful because they can be very easily integrated - just include the header, and build. No need to add another .c to the project, get its compile and link target inserted in the project. The single-.h approach makes sense for a project like this, where indeed the purpose is to put an embedded language/compiler, with as little fuss as possible, in any particular C-based project.
If the author caters to the desire to scratch the convention itch, and moves everything to .c - this just adds another layer of cyclomatic complexity to the project management. I'm a lot more inclined to try out such a radically useful project if its a single header - or can be expressed as a single header - than if I have to edit my project, add new .c targets, get the linking done properly, etc. Single .h with header guards - done properly (i.e. namespace) - can make such projects immediately more attractive than if it comes with a heavy load to integrate ..
This is false. A header guard will prevent a .h file being included multiple times inside a .c file. It will not prevent a .h file being included in multiple .c files because each .c file is its own compilation unit (well, usually at least). It doesn't even make sense preventing a .h file being included in multiple .c files because what purpose would a .h file even have then?
> Why would a competent developer use this .h multiple times?
Transitivity.
When you include a header file, it might include other header files. There are no real restrictions on what other header files it might include, as they may be part of the implementation, rather than the interface. We see this particularly clearly in C++, where private members and member-functions must be declared in header files.
The header files included by a header file, may change across versions.
Obviously, we don't want to omit any of our #includes just because a library we use happens to include it. This wouldn't solve our problem anyway, though: different headers may overlap in the headers they include.
This means it must be safe to include a header several times (i.e. inclusion of a header file must be idempotent). Diligence isn't enough. Fortunately, it's trivial to do this with the 'header guard' pattern (#ifndef MY_HEADER_H...)
Nice job. Read the first couple of entries, look forward to following through the rest. I enjoy this kind of exercise, just complex enough to give you a flavour of how the 'real' thing works whilst not being a mere toy example.
Recursive-descent which is LL is very common for parsing because of its simplicity; LR is theoretically more powerful, but in practice there doesn't seem to have been any great need for it. GCC, Clang/LLVM, MSVC, javac, csc (the C# one), and a whole bunch of other compilers use RD (often with some sort of precedence-climbing, which is essentially a refactoring of RD). GCC was originally using a generated LR parser (Bison) but moved to RD a long time ago.
23 comments
[ 1.8 ms ] story [ 63.1 ms ] threadThis is just a odd way of doing things. There are good reasons you don't usually see this, add one more .c file and things get awkward real fast.
Sure, it works in this particular instance, but it just makes me feel itchy. Maybe it's past trauma where I saw real weird codebases with static functions in .h files used across multiple .c files. It was just horrible to work with and reason about.
Hahah, yeah it smells, can't argue with that. I believe that everyone had a trauma like that in the past :)
Including one .c file in another is a bit odd, but less surprising. But then having include guards makes the entire thing a bit weird :)
B is a place that I want to use it, and it's tightly related to A.
Unfortunately, B belongs in a .c file.
But to use A in multiple places, it belongs in a .h file.
Crap, now these two very related things that I wanted to keep together lexically must now be split into separate files.
In this case, the author is making a super-simple example project. I'm happy they chose to forgo the standard C header/source file dance. It made it a lot easier to read and understand.
But yeah, the dance is required for "real" projects, as you've described.
- Put everything in the C file.
- If you need a function visible outside, put it's _prototype_ in the .h file.
- If you want to have a variable visible outside, also make a extern declaration inside the .h file.
- If you want to have anything else visible to the outside, just move it to the .h file.
The reason I say that putting functions inside a .h file is a pain is because of the way the preprocessor works. It takes the contents of the included file and pastes it at the place of the include. So doing a function inside a .h file and using it inside multiple .c files will mean you have the same function multiple times and it will throw a linker error if the function is not static. At link time the linker will be confused because it will not know about witch of the functions you're talking about when you call it. So then you declare it static, but then you _actually_ have 2x functions inside your executable, that do the same thing.
This is even worse with variables declared static inside .h files since here you will think you're changing your variable, but you actually have separate variables inside each compiled file.
Of course, my observation is also based on taste. I like when I have everything inside one file, instead of spread over multiple files( of course, as long as this is reasonable). This program here might have had one single .c file and it would have been just file. I actually would have considered it easier to read, but as I said, it's a matter of taste at this point.
This is prevented by a header guard.
And, also: if you have the same function defined elsewhere, and your compiler complains about it: good. That is a bug.
Single-header projects like this are useful because they can be very easily integrated - just include the header, and build. No need to add another .c to the project, get its compile and link target inserted in the project. The single-.h approach makes sense for a project like this, where indeed the purpose is to put an embedded language/compiler, with as little fuss as possible, in any particular C-based project.
If the author caters to the desire to scratch the convention itch, and moves everything to .c - this just adds another layer of cyclomatic complexity to the project management. I'm a lot more inclined to try out such a radically useful project if its a single header - or can be expressed as a single header - than if I have to edit my project, add new .c targets, get the linking done properly, etc. Single .h with header guards - done properly (i.e. namespace) - can make such projects immediately more attractive than if it comes with a heavy load to integrate ..
This is false. A header guard will prevent a .h file being included multiple times inside a .c file. It will not prevent a .h file being included in multiple .c files because each .c file is its own compilation unit (well, usually at least). It doesn't even make sense preventing a .h file being included in multiple .c files because what purpose would a .h file even have then?
These arguments hinge on the Dev being pretty careless.
This bridge doesn't need guard rails. A competent pedestrian wouldn't walk off the edge.
Transitivity.
When you include a header file, it might include other header files. There are no real restrictions on what other header files it might include, as they may be part of the implementation, rather than the interface. We see this particularly clearly in C++, where private members and member-functions must be declared in header files.
The header files included by a header file, may change across versions.
Obviously, we don't want to omit any of our #includes just because a library we use happens to include it. This wouldn't solve our problem anyway, though: different headers may overlap in the headers they include.
This means it must be safe to include a header several times (i.e. inclusion of a header file must be idempotent). Diligence isn't enough. Fortunately, it's trivial to do this with the 'header guard' pattern (#ifndef MY_HEADER_H...)
Without a doubt LR parsers prevail on the market. They tend to be more powerful & complex. Whereas LL parsers are easier to implement & understand.
That said, LL parsers are ideal for getting a bit of a taste of complicated compiler-related-stuff-world, in my opinion.