Here is a flagrant illustration of the robustness principle, or rather, of a failure to honor it.

I was investigating a bug in Declt where some floating point numbers were printed with exponent markers (e.g. 0.5f0 instead of just 0.5) in the Texinfo file, which broke the parsing of the file by Perl.

Eventually, I found out a double infringement of the robustness principle. First of all, Declt failed to comply with part 1 of the robustness principle: be lenient with the others. The Texinfo file generation routine should have been wrapped into a call to WITH-STANDARD-IO-SYNTAX and it wasn't. Always do that to be on the safe side. Lesson learnt.

This failure on my part, however, had the interesting consequence of exhibiting what I consider a serious infringement of part 2 of the robustness principle: be strict with yourself. It would have remained unnocited otherwise. The culprit here is not Declt. This time, it's the common-lisp-stat library. Problem: the simple fact of loading this library globally changes the value of *READ-DEFAULT-FLOAT-FORMAT* from SINGLE-FLOAT (the default) to DOUBLE-FLOAT. This is bad, and it can break your code in all sorts of nasty ways.

Explanation

*READ-DEFAULT-FLOAT-FORMAT* tells the reader how to read floats when no exponent marker is provided. By default, 0.5 will be read as a SINGLE-FLOAT. But this variable also influences the printer (out of a concern for printing readably I guess): when printing a float of a different format than the current default, then the appropriate exponent marker is also printed. So here is precisely what happened. Declt had been compiled with some floats (e.g. 0.5) read as SINGLE-FLOATs. Later on, those floats were supposed to be printed aesthetically as such. But right after loading common-lisp-stat, the default format changed to DOUBLE-FLOAT and all of a sudden 0.5 started to be printed as 0.5f0.

Consequences

This is bad enough already, but consider that messing with the standard IO syntax globally like this can break your code in all other sorts of even nastier ways. Imagine for instance that common-lisp-stat had been loaded before Declt, and Declt needed to be recompiled. All of a sudden, Declt would be using double floats and the bug would be gone. That is, until the next start of the REPL, after which all floats would be printed like 0.5d0!

So granted, my code wasn't robust enough. But please, don't mess with standard IO syntax globally!