Our main codebase used unix_conventional_names. Then, a GUI (FLTK, C++) was added, with classes like Fl_DoubleWindow; Every GUI part now had two conventions.

Then, it turned out that the abstractions FLTK gave on Windows needed a small nudge. So we added some Windows-specific stuff; that module had calls like FLTK Fl_RadioButton, Win32's TrackMouse, our main code's "user_action_t".

Then we added needed to add Python scripting (implementing a native Python module in C) - and since the most maintainable way to do that was to have the C names and Python names correspond, there's a module that also had Python convention names in it. Lather, rinse, repeat with JNI.

So, I would guess ~ 70% of the files ended up being "uniform convention" (with one convention, depending on source code language), and 30% had mixed conventions of up to 4 conventions.

And it didn't make anything a little bit unreadable. It's no more distracting than changing fonts in a document every paragraph (and occasionally in the middle of a paragraph) between several readable standard fonts (arial, courier, consolas, ...). It looks weird and bothering for the first few days, but is not actually distracting or hampering in any way.

And if anyone is going to reply "but new people who come into the codebase will be confused" - that codebase was moving millions of dollars per day, and was nontrivial enough that I wouldn't allow anyone to commit a change on their first week, often their first month, without two other people reviewing it (experienced people got only one person to review their code).

By the time anyone knew the code well enough to make a change, they weren't bothered by the multiple conventions either.

> doesn't make for pleasant reading, I'm sure you agree.

only lifespan, the upper case implying constantness when it isn't, bothers me. Other than that - pleasant as day.