Re: 15.0 LONG buildworld time

With these make flags:

WITH_META_MODE=yes
WITHOUT_CLEAN=yes
WITH_MALLOC_PRODUCTION=yes
WITH_CCACHE_BUILD=yes
WITHOUT_LLVM_ASSERTIONS=yes
WITHOUT_CLANG_BOOTSTRAP=yes

[rebuilding a freshly built world + kernel]

>>> World built in 106 seconds, ncpu: 16, make -j8
>>> Kernel(s)  TEST built in 18 seconds, ncpu: 16, make -j8

[TEST is GENERIC + options DDB]

After removing
WITHOUT_CLEAN=yes

>>> World built in 91 seconds, ncpu: 16, make -j8
>>> Kernel(s)  TEST built in 15 seconds, ncpu: 16, make -j8

[No idea why it is *less*!]

Build is on a 7 year old 14.3-STABLE ryzen 2700 zfs root system (which also
runs a couple of VMs).

> On Aug 15, 2025, at 2:37 PM, David Cross <david@crossfamilyweb.com> wrote:
> 
> I just upgraded one of my machines to 15.0 for testing some USB related work (see other thread), and one thing I noticed was that AFTER upgrading to 15.0 build times *doubled*.  This is unusual since the *first* build is typically the longest since it has to build the new toolchain before compiling, and after that it just uses the system build chain.
> 
> 
> In my case the build time (single threaded) was 38,000 seconds on 14.2, however after rebooting and doing a second build it shot to 76,000 seconds (about) -- I had rebuilt a second time since I thought the existence of a 14.2 library for SSL at build time contaminated the build, it didn't, but an error I got at startup from unbound made me think it did.
> 
> 
> It seems the GENERIC kernel in the 15.0 branch has some extra debugging and validation, but builds shouldn't be primarily kernel limited should they?  Build options are mostly stock, I add -O2, and REPTOLINE options, and set CPU type to my native (raptorlake).  Nothing else of note, and certainly have never had any other issues.
> 
>