Mac Lane's Idea of Structure versus Weil's

On Sat, May 26, 2012 at 7:48 PM, Michael Barr <barr <at> math.mcgill.ca> wrote:

> Let me point out that not every structure comes with an obvious notion of
> morphism.

This is entirely true on Bourbaki's theory of structures.  Mike gives
topological spaces as a good example.  Should open sets be preserved
by morphisms? or reflected? or both?

Each Bourbaki structure comes with a unique obvious notion of
isomorphism: a bijection which preserves and reflects all the data.
Each one comes with as many possible definitions of morphism as there
are ways to choose which data to preserve and which to reflect.  Weil
correctly understood this, as shown in the quote:

\begin{quotation} As you know, my honorable colleague Mac~Lane
maintains every notion of structure necessarily brings with it a
notion of homomorphism, which consists of indicating, for each of the
data that make up the structure, which ones behave covariantly and
which contravariantly [\dots] what do you think we can gain from this
kind of consideration? (Weil letter to Chevalley 1951).\end{quotation}

But Saunders was not thinking of any formal definition of "structure".
  He meant that in fact wherever you see mathematicians using some
notion of space or algebra, or whatever, you will see a notion of
homomorphism used with it.   And he was largely right, though he also
helped to make this more strictly true by convincing people it was a
useful perspective.

There was by then a well-established notion of topological space with