[Gretl-devel] Some basic info on the gretl source [was:Signalling]

On Tue, 8 Jan 2013, Allin Cottrell wrote: Roughly, here's how the C source is organised: the most important subdirectory is "lib/src", which contains the source to the dynamic library. Then you have "plugin", where you find all the stuff that is not contained in the library, but is used by the clients: mainly advanced statistical methods (like eg vecm or dpanel), but also other things, like the importers from foreign file formats. Then, in the "cli" and "gui2" directories, you have the source for the CLI and GUI clients, respectively. Inside the lib/src/ subdirectory, as Allin said, the first thing to look at is probably interact.c and the function gretl_cmd_exec(): the main idea of both the CLI and GUI clients is that they run a huge endless loop which examines user input and translates it into gretl "actions" as appropriate. So for example the command "nulldata", given at the CLI or GUI console, or the action "File > New data set" in the GUI client translate into the NULLDATA internal gretl command. You'll find a complete list of such commands in the file lib/src/gretl_commands.h. Of course, the most important command of all is "genr", so there's a good deal of C code which deals exclusively with it. If you want to have a look at how things are done in that department, be aware that it's one of the most complicated parts of the gretl source, and it's not useless to have an idea of how the main techniques for lexing/parsing/evaluating expressions work in compilers. Basically, however, the "lexer" (genlex.c) translates a string like, say, "scalar a = exp(2+1)" into a "parse tree", which is then "evaluated" by the eval() function (geneval.c), which is typically called recursively and calls the number-crunching functions (which may be in geneval.c itself, or, in other cases, in dedicated files like genfuncs.c). As for estimation techniques, something that is probably useful to get you started is that practically all of them work by filling up a container called a MODEL. The description of what a MODEL is can be found in lib/src/libgretl.h at about line 300, but suffice it to say that it's not unlike a hansl bundle, that you fill up with numbers as you compute them. So for example, when you estimate a probit model, you first set up an empty MODEL, then you fill up the "coeff" array with the estimated coefficients, etcetera. Once the container is properly filled with all the good stuff, it gets printed and is put somewhere (the "stack") where the client program can find it if needed. Actually, probit/logit models are a nice place to start to get the hang of how things are typically done: you'll find those in the function binary_model() in lib/src/discrete.c (don't start from OLS: the code in there is decidedly harder to follow; but if you have to, look at ar1_lsq() in lib/src/estimate.c). Sorry for the long message, but you asked for it! :) ------------------------------------------------------- Riccardo (Jack) Lucchetti Dipartimento di Scienze Economiche e Sociali (DiSES) Università Politecnica delle Marche (formerly known as Università di Ancona) r.lucchetti(a)univpm.it http://www2.econ.univpm.it/servizi/hpp/lucchetti -------------------------------------------------------