On Tue Nov 17, 2026 at 02:24:20AM +0100, Kirill A. Korinsky wrote:
> On Sat, 17 Oct 2026 23:59:24 +0200,
> Kirill A. Korinsky <[email protected]> wrote:
> > 
> > On Thu, 15 Oct 2026 12:44:23 +0200,
> > Theo Buehler <[email protected]> wrote:
> > > 
> > > > Anyway, regarding the size: it is large but without it modern lean is
> > > > somehow useless. Almost everything inside mathlib.
> > > 
> > > I know what you mean.
> > > 
> > > You have my approval for the update. Not mathlib. I gave my reasons for
> > > suggesting to remove lean a while back.
> > > 
> > 
> > I have other approach.
> > 
> > Instead build it by hand I can get .olean (and other files) from official
> > mathlib cache.
> > 
> > I see two approches:
> > 1. download everything in by ports;
> > 2. prepare repacked cache as archive.
> > 
> > Because it ends with fetching near of 10k files with name as it's hash and
> > building custom tool to work with it...  I think (2) is the right solution.
> > 
> > I had made https://github.com/catap/mathlib-cache/ to be transparent how
> > that archive is made.
> > 
> > Why do we need it?
> > 
> > Right now user can use lake and add mathlib as dependency but it will
> > downloads all files into .lake in the project which needs ~10G.
> > 
> > I think that is not that usefull.
> > 
> > Ship sources and compile on demand is also wrong, because
> > 
> > import Mathlib
> > 
> > will end with compiling everything (plus it needs the same ~10G at .lake)
> > 
> > So, here a solution which uses repacked .olake and friends from official 
> > cache
> > 
> 
> Lean and mathlib was updated to 4.36.0
> 
> Here updated mathlib port which builds fast.
> 
> Anyone?
> 
> -- 
> wbr, Kirill

Package fine, expect form the fact that this distfile is a real stress
test for my disk, OK rsadowski@

Reply via email to