On Tue, 2 Feb 2016, Barry Smith wrote:

> 
> > On Feb 2, 2016, at 10:01 PM, Jed Brown <[email protected]> wrote:
> > 
> > Barry Smith <[email protected]> writes:
> > 
> >>> On Feb 2, 2016, at 7:00 PM, Jed Brown <[email protected]> wrote:
> >>> 
> >>> Satish Balay <[email protected]> writes:
> >>>> Or you can copy the whole '$PETSC_ARCH/externalpackages' dir over to
> >>>> this other machine [and place in the same location - for any
> >>>> PETSC_ARCH you plan to use..]
> >>> 
> >>> This sounds unreliable.
> >> 
> >>  In what way? 
> > 
> > You'll be building in a dirty directory (one that was used for a
> > different PETSC_ARCH).  If the project doesn't manage their dependencies
> > perfectly, it could use a build artifact from a different PETSC_ARCH,
> > which would result in linking errors (for example).
> 
>   You are right, this could be an issue

Well my recommendation was for this use case of "same install on a different 
machine"

And currently we do kinda support inplace repeated install of the same
package

[otherwise configure should be deleting all sources after build - and
redownload each time. We don't do that]

And even if the PETSC_ARCH1/externalpackages is copied over to
PETSC_ARCH2/externalpackages - builds should work. the primary reason
we moved externalpackages from PETSC_DIR/externalpackages to
PETSC_DIR/PETSC_ARCH/externalpackages is to support concurrent builds
in these locations.

The issues you raise come about with using incompatible versions of
packages.  This can be due to switching git branches - or rebuilding
after 'git pull' - that might have changed package dependencies.. But
thats not the use case - as I understood it..

Sure - keeping track of all the tarballs for a given petsc version is
a better approach.

> 
> >  I'd rather cache
> > the tarballs (with checksum) or use git repositories (which can be
> > reliably cleaned)

I sure would like a common cache [for any build/arch/version] for both
tarballs and git repos - so network traffic can be cut down. [say in
~/.local/petsc/cache].

> 
>   Ok then you write the code to do this. It doesn't currently exist (and I am 
> not sure we want to be in the business of writing a full package management 
> system :-).
> 
>    Anybody who thinks that a computer system that doesn't allow direct 
> downloads of open source tarballs or git repositories (because of security 
> considerations) but allows people to indirectly download those exact same 
> tarballs or git repositories (as if that indirect download somehow magically 
> cleanses the tarballs or repositories) is a damn fool anyways.

This is an easy way to cut-down syadmin overhead [or system
complexity/downtime] for special-use systems. For ex; JLSE has various
types of hardware. Its not easy to make sure they are secure enough
for all network traffic. So all of them [except front-end general-use
linux boxes] are firewalled off.

Other systems provide proxies [to monitor all such traffic] - This is
not easy to use via configure. [even though python urllib supports
proxies..]

Satish

> 
>   Barry
> 
> > than simply build in dirty directories that cannot be
> > cleaned.
> 
> 

Reply via email to