Summary: | dev-db/mongodb-5.0.5-r2 fails to build, warning that AVX is required, even when USE=cpu_flags_x86_avx is OFF | ||
---|---|---|---|
Product: | Gentoo Linux | Reporter: | Phil Stracchino (Unix Ronin) <phils> |
Component: | Current packages | Assignee: | Ultrabug <ultrabug> |
Status: | RESOLVED FIXED | ||
Severity: | normal | CC: | jah, phils |
Priority: | Normal | ||
Version: | unspecified | ||
Hardware: | All | ||
OS: | Linux | ||
See Also: |
https://bugs.gentoo.org/show_bug.cgi?id=809692 https://bugs.gentoo.org/show_bug.cgi?id=908987 |
||
Whiteboard: | |||
Package list: | Runtime testing required: | --- | |
Attachments: | Allow mongodb to be build without AVX support |
Description
Phil Stracchino (Unix Ronin)
2023-01-09 15:30:23 UTC
I think the non-use of REQUIRED_USE was deliberate here to give a more informative message. MongoDB 5.x doesn't work without AVX on amd64. What do you want us to do instead? (In reply to Sam James from comment #1) > I think the non-use of REQUIRED_USE was deliberate here to give a more > informative message. > > MongoDB 5.x doesn't work without AVX on amd64. > > What do you want us to do instead? Correct me if I'm wrong: You're saying that, USE flag notwithstanding, AVX is a non-negotiable hard requirement in MongoDB 5? (In reply to Phil Stracchino (Unix Ronin) from comment #2) > (In reply to Sam James from comment #1) > > I think the non-use of REQUIRED_USE was deliberate here to give a more > > informative message. > > > > MongoDB 5.x doesn't work without AVX on amd64. > > > > What do you want us to do instead? > > Correct me if I'm wrong: You're saying that, USE flag notwithstanding, AVX > is a non-negotiable hard requirement in MongoDB 5? Yes, as unreasonable as that is :( See https://www.mongodb.com/docs/v5.0/administration/production-notes/#platform-support-notes and bug 809692. I'd missed Piotr's comment on that latter bug, so maybe there's some hope for us to implement a hack. (In reply to Sam James from comment #3) > (In reply to Phil Stracchino (Unix Ronin) from comment #2) > > (In reply to Sam James from comment #1) > > > I think the non-use of REQUIRED_USE was deliberate here to give a more > > > informative message. > > > > > > MongoDB 5.x doesn't work without AVX on amd64. > > > > > > What do you want us to do instead? > > > > Correct me if I'm wrong: You're saying that, USE flag notwithstanding, AVX > > is a non-negotiable hard requirement in MongoDB 5? > > Yes, as unreasonable as that is :( > > See > https://www.mongodb.com/docs/v5.0/administration/production-notes/#platform- > support-notes and bug 809692. I'd missed Piotr's comment on that latter bug, > so maybe there's some hope for us to implement a hack. In that case my apologies for misunderstanding, this is not a bug against the ebuild, this is "working as upstream intended", however questionable upstream's decision may be. I for one don't have the option of just arbitrarily replacing two 1U rack servers just to support mongodb 5. (The only reason I'm using mongodb in the first place is net-wireless/unifi.) Created attachment 848275 [details, diff]
Allow mongodb to be build without AVX support
AVX in mongodb 5 is required because by default they use -march=sandybridge.
You can disable this by passing --experimental-optimization=-sandybridge to SCons.
This patch to the ebuild does just that. I was able to build mongo and use it on my Celeron J1900.
(In reply to Guy Martin from comment #6) > Created attachment 848275 [details, diff] [details, diff] > Allow mongodb to be build without AVX support > > AVX in mongodb 5 is required because by default they use -march=sandybridge. > You can disable this by passing --experimental-optimization=-sandybridge to > SCons. > > This patch to the ebuild does just that. I was able to build mongo and use > it on my Celeron J1900. Nice find sir! > > Yes, as unreasonable as that is :(
> >
> > See
> > https://www.mongodb.com/docs/v5.0/administration/production-notes/#platform-
> > support-notes and bug 809692. I'd missed Piotr's comment on that latter bug,
> > so maybe there's some hope for us to implement a hack.
>
> In that case my apologies for misunderstanding, this is not a bug against
> the ebuild, this is "working as upstream intended", however questionable
> upstream's decision may be. I for one don't have the option of just
> arbitrarily replacing two 1U rack servers just to support mongodb 5.
>
> (The only reason I'm using mongodb in the first place is net-wireless/unifi.)
+1 ... virtual environment. Hard-masked >=5.0.0 for the moment.
I'm gonna add this feature with an added WARNING instead of silently enabling the experimental scons option on the upcoming 5.0.16 ebuild bump fixed in 5.0.16 |