FPKM computed on gene effective length or annotation length?

[bhdobrin]

Hello, I'm interpreting someone else's results and I'd like to understand how FPKM is computed in stringtie. I have searched the code and found likely relevant variables like Frag_Len, calc_fpkm, tcov and the like, but there is a lot of code there to sort though! Also I searched the issues for "effective length" in case the question had been asked, but there were no hits. FPKM values will be more conservative (i.e., slightly underestimate the true expression) if computed on annotation length, which is longer, so I'd like to know which figure I'm dealing with. Thanks very much!