Use sg lightweight instead of igraph

[gaborcsardi]

If you are interested, I will provide a PR soonish. Either for switching to https://github.com/mangothecat/simplegraph which is R-only and has no dependencies, or custom code. It is fairly trivial.

[jimhester]

Sounds good to me, I had initially used igraph as a performance improvement, but it is a heavy dependency for sure.

[gaborcsardi]

Do you have an idea how big the graphs are?

[jimhester]

Depends on the file you are linting I guess, I was testing on https://github.com/wch/r-source/blob/trunk/src/library/tools/R/QC.R which is ~8300 lines long. Lintr splits the parse data data frames by expression, looks like the largest dataframe is 6652 rows, which I guess is fairly small...

[gaborcsardi]

Indeed, my quick and dirty R implementation will not do. For QC.R it increases running time from 70 sec to 223 sec. https://github.com/gaborcsardi/lintr/commit/7121cf8d86b547539f7ebc773910bd05667eccac

I'll try sg else in a minute.

Btw. is this graph component stuff only needed for splitting the file into expressions? Because then maybe I can ~~~start~~~ try sg else.

Btw. 2. it is easy to write this in C as well, is that preferable to igraph? I am not sure myself. Maybe the best would be some micro-package dependency that only does this, but fast (= using C code).

[gaborcsardi]

OK, the bottleneck was a stupid type conversion (did you know that modifyList is super slow? It makes sense if you look at its code...), and now it is actually faster than the current lintr code, it is ~61 sec for QC.R. Hopefully it is also correct. :) Let me run some more tests, and then I'll submit a PR.