Change: move GET iterator time conversion out of SQL

[mattmundell]

What

Move the ISO time calculations in the "GET" iterators from SQL to C.

This is for the columns creation_time and modification_time.

Note that plain C iso_time works the same as SQL iso_time (creation_time, opts.user_zone) because the C does a tzset before the iso_time. To test this use GMP like:
time o m m '<get_reports report_id="82f3531c-43cd-4453-8300-990dc423d96b" delta_report_id="ea5d9e94-4f48-4a16-b0b5-a63b637176d2" details="1" filter="rows=2 timezone=UTC"/>'

Also, removal of opt.user_zone is safe from a filtering perspective, because for the creation_time/created filter term is converted to an epoch.

Why

Having the SQL do the time conversion is slower when there are many rows with a large offset, because Postgres still does the time conversion for every row.

time o m m '<get_info type="nvt" filter="sort-reverse=created rows=10 first=80000"/>' > /tmp/info
patch: 0.3, 0.3, 0.3
main: 3.2, 3.2, 3.3

This PR leads to faster SecInfo pages in GSA when paging to the end, CPEs in particular.

References

https://www.postgresql.org/docs/current/queries-limit.html 7.6 LIMIT and OFFSET:
The rows skipped by an OFFSET clause still have to be computed inside the server; therefore a large OFFSET might be inefficient.

[github-actions]

Conventional Commits Report

Type	Number
Changed	1

🚀 Conventional commits found.

[timopollmeier]

This looks like a good optimization but there is one thing I'd do differently:

I think it would be better to make the allocation of new strings that need to be freed by the caller more explicit.

My suggestion would be to return an int64 from the iterator function and pass that to a new function that does both the ISO date-time conversion and the allocation.

[mattmundell]

I think it would be better to make the allocation of new strings that need to be freed by the caller more explicit.

My suggestion would be to return an int64 from the iterator function and pass that to a new function that does both the ISO date-time conversion and the allocation.

Good idea. I've started with 1e1e147. Will get back to it next week.

[mattmundell]

I think the clang-format config should change to allow breaks after the return type of short definitions.

[mattmundell]

I think the clang-format config should change to allow breaks after the return type of short definitions.

Hmm, .clang-format already has AlwaysBreakAfterReturnType All.

[mattmundell]

Hmm, .clang-format already has AlwaysBreakAfterReturnType All.

Due to a clang-format limitation, see llvm/llvm-project#76206 (comment). Worked around in cb983df.

[a-h-abdelsalam]

Looks good! You may want to update report_configs as well after updating the branch.

[mattmundell]

Looks good! You may want to update report_configs as well after updating the branch.

Thanks for going through this big PR. Report configs part done in be66e2e.

Requested changes have been made