Deprecate SPT threading support on NonStop.

[rsbeckerca]

This fix removes explicit support for the SPT threading model in configurations. This also reverts commit f63e1b4 that were required for SPT but broke other models.

Checklist

• documentation is added or updated

[rsbeckerca]

This pull request has been tested on all threading and unthreading models on NonStop.

[hlandau]

Some minor wording quibbles, but otherwise looks good.

[t8m]

I'd consider splitting this into two commits/PRs. One that reverts the bio_sock change as that should go to 3.2 as well, perhaps mentioning that QUIC might not fully work on SPT. And the other that should go into master branch that removes it completely?

[rsbeckerca]

I'd consider splitting this into two commits/PRs. One that reverts the bio_sock change as that should go to 3.2 as well, perhaps mentioning that QUIC might not fully work on SPT. And the other that should go into master branch that removes it completely?

Both changes need to go to both branches. Do you mean two PRs that do the same thing but for two branches?

[rsbeckerca]

I'd consider splitting this into two commits/PRs. One that reverts the bio_sock change as that should go to 3.2 as well, perhaps mentioning that QUIC might not fully work on SPT. And the other that should go into master branch that removes it completely?

We also determined that any use of SPT on NonStop would break OpenSSL because of the BIO changes. That model had to be dropped forever. 3.1 and 3.0 are fine because DGRAM is not used, which is where SPT falls down.

[t8m]

If you build 3.2 with no-quic the new things in BIO_dgram won't be used. It looks fairly strange to remove compilation targets on already released stable version. Perhaps the correct fix would be to disable quic for SPT targets automatically on 3.2?

[rsbeckerca]

If you build 3.2 with no-quic the new things in BIO_dgram won't be used. It looks fairly strange to remove compilation targets on already released stable version. Perhaps the correct fix would be to disable quic for SPT targets automatically on 3.2?

Wait... 3.2 is not officially release. We are keeping SPT on 3.0 and 3.1. That does not change. This is for 3.2 and master. The reason it does not work for 3.2 is that the DGRAM code requires a part of SPT that enables new processing on fgets that does not work. spt_fgetsx, which is what is used at 3.2 to get BIO to compile with SPT, causes a missed EOF detection on configuration files regardless of no-quic. SPT cannot be part of any 3.2 build at all, regardless of the options selected otherwise it hangs across the board reading openssl.cnf.

[t8m]

3.2.0 was released today. I understand that you did not release the binaries for NonStop yet but there are no changes going into 3.2.0. As for the source code base the 3.2 is a stable branch and there shouldn't be added changes like removing build targets. I understand that without reverting f63e1b4 the 3.2 is totally unusable for NonStop but that is a downstream change and something that we certainly want to apply upstream before the 3.2.1 release.

[rsbeckerca]

3.2.0 was released today. I understand that you did not release the binaries for NonStop yet but there are no changes going into 3.2.0. As for the source code base the 3.2 is a stable branch and there shouldn't be added changes like removing build targets. I understand that without reverting f63e1b4 the 3.2 is totally unusable for NonStop but that is a downstream change and something that we certainly want to apply upstream before the 3.2.1 release.

That is correct. We cannot and will not release 3.2.0 binaries. I wish that was in the notes for it. I will try to tell the entire community that 3.2.0 is unusable.

[t8m]

IMO this deserves an OTC discussion.

OTC: Is it acceptable to remove the NonStop SPT build targets in 3.2 branch? QUIC cannot be properly supported on SPT.

[hlandau]

@rsbeckerca Can you link any NonStop documentation on the different threading models and their merits/uses?

Do you have any data or rough feeling on usage levels of these different threading models?

[beldmit]

Do we have to remove this support? Could we just enforce disable-quic for this option?

[rsbeckerca]

@rsbeckerca Can you link any NonStop documentation on the different threading models and their merits/uses?

Do you have any data or rough feeling on usage levels of these different threading models?

I have no evidence that anyone is using 3.2 in SPT. There was only one download in our history of SPT for 1.1.1. We know of one company - my personal contacts - who used OpenSSL it, but have since moved to PUT.

[rsbeckerca]

Do we have to remove this support? Could we just enforce disable-quic for this option?

Simply building with SPT will trigger the problem because fgets() is replaced with spt_fgetsx(), which hangs reading openssl.cnf regardless of whether QUIC is used. We are highly unlikely to get fixes to SPT at this stage for the problem for the specific problem from the platform vendor because of the limited support state of this threading model. With no evidence of anyone using OpenSSL with SPT on NonStop - or at least support of 3.0 and 3.1 with SPT through those series end of life, customers have options if needed.

[t8m]

Simply building with SPT will trigger the problem because fgets() is replaced with spt_fgetsx(), which hangs reading openssl.cnf regardless of whether QUIC is used.

That can be fixed by reverting f63e1b4 and faa4351

[rsbeckerca]

Simply building with SPT will trigger the problem because fgets() is replaced with spt_fgetsx(), which hangs reading openssl.cnf regardless of whether QUIC is used.

That can be fixed by reverting f63e1b4 and faa4351

I agree. I can revise the PR if OpenSSL wants to keep the SPT configurations although I will no longer be building or maintaining those going forward.

[openssl-machine]

24 hours has passed since 'approval: done' was set, but as this PR has been updated in that time the label 'approval: ready to merge' is not being automatically set. Please review the updates and set the label manually.

[nhorman]

@rsbeckerca Just fyi, the OTC met today and held a vote on this issue, decision was:

1. Remove config target from master immediately (no process question there)
2. Remove config target from 3.2 after announcing its removal on our various communication channels, just to give current users of a stable branch to voice concerns prior to said removal.

Thanks for your work on this!

[hlandau]

From the minutes:

    topic: Support for the NonStop SPT threading model and SPT build configs
           can be removed from 3.2 branch subject to no objections being
           received from the community within two weeks after this vote passing
    accepted:  yes  (for: 9, against: 0, abstained: 0, not voted: 1)

openssl/technical-policies#86

[rsbeckerca]

From the minutes:

    topic: Support for the NonStop SPT threading model and SPT build configs
           can be removed from 3.2 branch subject to no objections being
           received from the community within two weeks after this vote passing
    accepted:  yes  (for: 9, against: 0, abstained: 0, not voted: 1)

openssl/technical-policies#86

Thank you for your support in this matter.

[nhorman]

@rsbeckerca the two week timer on public comments has expired on this, feel free to merge it, or let me know and I can take care of it

[rsbeckerca]

@rsbeckerca the two week timer on public comments has expired on this, feel free to merge it, or let me know and I can take care of it

I would love to have the authority to merge on this project, but do not. Would you be able to handle it?

[nhorman]

ack

[nhorman]

merged to master and 3.2

[rsbeckerca]

@nhorman Thank you