What are the dusty corners a newcomer to CMake wil

问题:

I've done a lot of projects and a lot of different build systems and CI tools. Most recently, I've been exposed to the occasionally challenging task of adding to an autotools based environment for a reasonably sized C++ application. While I love the ease of use for the end user, I'm not so fond of dealing with m4 and all of the auto* tools from the developer side.

I'm working on a reasonably large side project in my spare time and have decided that I'd like to take CMake for a test drive. Since I'm just starting out, I'm obviously planning on digging through the documentation, FAQ, wikis, etc. and learning by doing. BTW, I'd fork over money for the "Mastering CMake" book, but the comments that I found on Amazon were enough to make me decide that it probably isn't worth the money. All that being said, in anything new there are often "gotchas" that a newcomer will often stumble over that the old pros have long since learned to avoid. I'm wondering what those are based on people's experience with CMake, and I'm hoping to cut my learning pains down a bit by asking here.

I should point out that I'm planning on building on Linux primarily, and other UN*X variants. Windows isn't really a concern from my POV. This is a large server-side application, with a Web and CLI interface for operators, a northbound REST interface for automation/integration with OSS tooling, and a SOAP southbound interface for CPEs. I'm going to need a lot of third-party libraries and applications to get this all to work unless I want to take the next 10 years to build this all by hand. :)

回答1:

First of all, I think CMake is an excellent build-tool. It has in my opinion by far the best support for multi-platform-builds and has a powerful mechanism for finding 3rd-party libraries. Combined with CPack it even provides reasonable options for packaging and installation. Some hints and possible problems:

1. Always aim for out-of-source builds: An obvious one: It can be difficult for a project designed for in-source builds to get it built out-of-source. So, if you can design it from the start, aim for out-of-source builds.

2. Syntax: The syntax can be bizarre with a lot of strange quirks, although this is getting better since the 2.6 and 2.8 versions.

3. Caching of variables. CMake keeps track of a cache of variables and settings and sometimes this can be a problem when rebuilding something. Try to remove the CMakeCache.txt (or clean your out-of-source build-directory) and rebuild. This one is also mentioned by DarenW.

4. Finding and configuring 3rd-party libraries: If you have a large project depending on several libraries (your own or 3rd-party), this will likely be the most troublesome.

   • I seriously recommend diving into the find_package command. It is very powerful, but fully depends on the quality of the FindXXX.cmake files. Especially when building on multiple platforms (mostly Windows and Mac), these files may require some tweaking or you may need to write your own.

   • Problems with linking libraries like debugging "undefined references" or mismatches between debug|release or static|shared can be difficult to debug. Especially with 3rd-party libraries these problems may be caused by incorrect 3rd-party FindXXX.cmake files...

回答2:

One thing that took me a while to figure out: if anything goes haywire with a build because I've changed a compiler option, added/removed a .c file, had confusion with support library versions, etc. - it's best to delete the build tree and run CMake from scratch. Otherwise, even though CMake and make seem to do their job, the resulting executable crashes.

It could be we're "not doing it right", but that's becoming a common problem with many tools, languages, frameworks, etc. It's not possible for everyone to be expert at all the software they depend on, and "not doing it right" is the only way a lot of things get done even among professionals. CMake is pretty smart about a lot of things, but not everything. Nuking the build tree remains a commonly used technique on our project.