Consider a package B. It has a dependency of package A. Package A is already installed. Package B fails to compile, citing a lack of package A as the reason. But A is there.

Note: both packages are in the book. Package A was installed following the book exactly. Package B's attempted compilation was also done as per the book. Md5sum's checked out.

So, what does one do in this situation?

Thanks you in advance.

p.s. I've had this happen in every distro and the various solutions (specifying --cpu, --mcpu, --arch, --march, sym-linking .so files like a fiend, etc.) never seemed to work.

I guess that package B is searching for package A in a wrong place... OR package A has some suffix not recognized by package A.
For e.g.:
- package B expects package A to be in /usr but it is in /opt or /usr/local
- package A has a suffix like '-qt5', being named as packageA-qt5, but package b searches for packageA

Best regards.

It might help to name packages, for 1 to make sure t installed ok, 2 if there is a problem it could be reported, 3 to help other people if they suffer the same.

Normally, my questions here are highly specific. This thread is more general, as it is about a problematic situation that I have run across in several distros over many years, which has never been resolved. Technically one could say this is not a Linux From Scratch question, however, because users of [B]LFS deal with compiling from source for everything, I thought here would be where the expertise for this sort of thing would reside.

Usually a google search of the error gives a clue as to what went wrong, as often someone else has run into the same situation.

Sometimes its a bug in package A.

Sometimes package A depends on package C, and fails to complain loud enough.

Sometimes package A needs to be configured with a particular command, before running make, so that package B can compile.

A lot of times there's a dependency that did not get mentioned because it was assumed that everyone would already have that dependency.

Wayne Sallee
Wayne@WayneSallee.com
http://www.WayneSallee.com

You appear to be looking for information on the Computer Science aspect of this. The terms you might find helpful are: circular dependency, dependency hell, mutual recursion and cyclic references. There are techniques to combat it ADP, which is requires revisiting code, like debugging, and programmers don't like to fix their mistakes until it browbeaten about them. The other is merging, which everyone now does because CS in the post OO era has forgotten important principals. The words "feature complete" are lost on an entire generation of software engineers.

Come to think of it, there's another method to combat it: runtime integration. But the shortest path to "putting out something cool" is to simply require it and move on, rather than check for the required library and report that the feature is unavailable. (Example: Go look at UDisks 2.7 in the current SVN and see what upstream did with libblockdev.)