On 5/20/19 9:58 PM, Andres Freund wrote: > Hi Andrew, > > On 2019-03-30 16:42:16 -0400, Andrew Dunstan wrote: >> On some machines (*cough* Mingw *cough*) installs are very slow. We've >> ameliorated this by allowing temp installs to be reused, but the >> pg_upgrade Makefile never got the message. Here's a patch that does >> that. I'd like to backpatch it, at least to 9.5 where we switched the >> pg_upgrade location. The risk seems appropriately low and it only >> affects our test regime. > I'm confused as to why this was done as a purely optional path, rather > than just ripping out the pg_upgrade specific install? > > See also discussion around > https://www.postgresql.org/message-id/21766.1558397960%40sss.pgh.pa.us >
By specifying NO_TEMP_INSTALL you are in effect certifying that there is already a suitable temp install available. But that might well not be the case. In fact, there have been several iterations of code to get the buildfarm client to check reasonable reliably that there is such an install before it chooses to use the flag. Note that the buildfarm doesn't run "make check-world" for reasons I have explained in the past. NO_TEMP_INSTALL is particularly valuable in saving time when running the TAP tests, especially on Mingw. cheers andrew -- Andrew Dunstan https://www.2ndQuadrant.com PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services