Anything with an "edit and continue" type feature I've found will sometimes miss-patch the partial build. It's rather flaky as to when I've seen this type of thing happen, but it's very annoying when it does.

in my experience, a seg-fault cropping up/disappearing with a clean is resonate of a much deeper problem in the code base.

This is not to say it very well might be an issue with a bad build, but in general, i've almost always discovered their to be an actual flaw in my code somewhere after heavy analyzing(something that only cropped up because the compiler didn't null all my pointers(and i forgot to explicity set a pointer to null) for me for some reason, or was allowing buffer-overflow's that didn't cause immidiate crash's)

edit after L. Spiro's/Karsten_'s posts, i'd like to point out that i only use a single machine for compilation, and i use VS 2010, which auto-detects header changes, and appropriately re-compiles the obj's that include the headers, which is probably why i more often experience such things in form of actual bugs, than bad build's.

That’s not been my experience (to slicer4ever).

Doing a clean build removes .OBJ files from different versions of compilers and this can have a profound effect on the stability of the resulting .EXE.

I even have the same compiler on both my Windows XP and Windows 7 installs, same service packs, etc. But when I switch between Windows XP and Windows 7 I really have to do a clean or I am guaranteed to get segment faults and other strange behavior.

Many times in the most isolated areas I have been able to prove the code was correct. A clean rebuild became part of my routine and I no longer suspect any underlying issues with the code.

As for the original poster, if you work with your code between different machines, save yourself some hassle and get into the habit of cleaning each time you copy the code and project to a new machine. Once again, for me, failure to do so always results in some kind of crazy run-time crash that makes no sense.

L. Spiro

Usually IDEs deal with this sort of thing using .d files.

Basically when you change a header and add something like an std::string to it, you really should compile every unit (.cpp file) that has a dependence on that header.

I believe this is because otherwise the space for the new variable is not reserved properly from those objects.

I have this issue when using Makefiles but rather than fix the issue (making the build system quite a bit more complex), I just detect whether a header has changed and rebuild the whole thing.

I'm with LSpiro here, have frequently had issues that were nothing to do with code and simply due to the IDE/Build System failing to rebuild an object file that it should have.

Qt's PRO system suffers from this quite often. Guess it is hard to get a dependency system right.