Opening a separate issue to track this, xref #1783.

I made some modifications to the script under travis to not compile fst binaries. That shaved a few minutes off it, but this is not an entire solution.

I tried to use ccache, which reduced the make all time from 6:25 (with ccache enabled and empty cache) to 0:47 (cache warm) on my machine, absent any source changes. I hoped with the cache warm the build will take a shorter time, but the catch 22 is travis does not save the cache directory when a run times out, so the next compile starts with the blank slate. I'll see if it is possible to force travis to push the cache explicitly after compile but before running tests.

Another thing I am going to try is skip linking the binaries (compile only). I can think of bugs that would slip through the holes this way, but there are potential much worse ones that would not be caught even by linking anyway.

Other ideas welcome.

/cc @dogancan @jtrmal