I used Octave for a year when my institution's Matlab license servers were being improperly administered. (I had a lot of project code written in Matlab, but the license server going down on the weekend before a conference deadline with nobody available to reboot it until Monday was a dealbreaker.) The biggest stumbling block was that Matlab has a huge and heavily used proprietary package library, and a lot of my existing code, (official) tutorials and Stack Overflow code assumed these libraries were available. In Octave I found myself reimplementing the newer parts of the Matlab image processing libraries. This led to the discovery that the Matlab and Octave builtins for handling image data are subtly different, so I ended up having to run tests in the code and write different conditional flows to make it cross-compatible. There are also subtle differences in basic behaviors (was it variable scoping? file handling?) which resulted in some surprise and frustration.

Following the licensing and Octave debacle, all my latest code is written in numpy.