> A grammer that has shift/reduce conflicts is poorly written.

Well, now ...

Since yacc specifies exactly what it does with a shift-reduce conflict
(always choose the shift), it is possible to have a grammar with 
conflicts where the author has reviewed all the conflicts reported
and confirmed that the parser will unambiguously do exactly what's intended.

Usually then there will be some comments in the documentation or makefile
saying "expect 3 reduce-reduce and 5 shift-reduce conflicts from this grammar"
and if yacc reports exactly that many, it's doing the right thing.

Why don't all grammar writers go further and continue refactoring their
grammars until the LALR parser generator doesn't report any conflicts?
Because the kind of refactoring needed to get rid of those last conflicts
can be quite elaborate and actually make the grammar longer, harder to read,
harder to understand, and harder to modify.  There are diminishing returns
when you cross the line from refactoring driven by the clarity of your
grammar to refactoring driven by the quirks of a parser-generator algorithm.
That's why generators like yacc offer reasonable defaults for common
ambiguities, and why some grammar writers don't mind taking advantage of them.

-Chap