+1 to this! Thank you for fixing it!

This commit needs further explanation and references, such as

This fixes ??? (I still don't understand why it is being restored)

Prior to this change DEBUGINFOD_URLS was disabled always since it was slowing down tests.

Ref: #8805

It is being "restored" because it unconditionally disables a valgrind feature. If you do not want to use this feature, then why have this environment variable set in the first place? As was shown, this feature is usable to some. I think that's motivation enough.

This fixes ???

Without repeating many posts in #8805 ...

Some distributions - like ArchLinux - don't distribute debug packages.
Instead they offer it though "debuginfod", a daemon to fetch debug symbols from a central service.

Without debug symbols valgrind errors (see #8805, the message has been posted there a couple times).
And that error causes nearly all curl tests to fail (again, only with valgrind installed).
With debug symbols (and in arch' case that means with DEBUGINFOD_URLS) these tests work just fine.

As was shown, this feature is usable to some. I think that's motivation enough.

I wasn't for or against it. It does need further explanation, such as...

Without debug symbols valgrind errors (see #8805, the message has been posted there a couple times).
And that error causes nearly all curl tests to fail (again, only with valgrind installed).
With debug symbols (and in arch' case that means with DEBUGINFOD_URLS) these tests work just fine.

Ok then may I suggest:

Prior to this change DEBUGINFOD_URLS was always disabled due to a report of it slowing down tests. However some distros need it to fetch debug symbols, and if it is disabled on those systems then curl tests with valgrind installed will fail.

Ref: #8805

etc

Without debug symbols valgrind errors

IMO this is a distro-specific problem: valgrind natively works very well without symbols. Otherwise how would you use it on a foreign stripped program ?
Each stock dtstro brings its own lot of problems: Arch forces you to use debuginfod, while Fedora works without but presets DEBUGINFOD_URLS at each login.

A solution for those who want to always disable it is to add a export DEBUGINFOD_URLS= line in ~/.profile.

In any case, if enabled and the full debug info for libc, ssl lib, ldap lib, etc is downloaded on a slow connection, be prepared to have timeouts at first test... And you'll have plenty of time for some beerscoffees !

@jay

Ok then may I suggest:

Prior to this change DEBUGINFOD_URLS was always disabled due to a report of it slowing down tests. However some distros need it to fetch debug symbols, and if it is disabled on those systems then curl tests with valgrind installed will fail.

I don't see a suggestion there except stating how it works prior to this patch.
Or is your suggestion to keep it as it was? (thus broken on Arch)

I don't see a suggestion there except stating how it works prior to this patch.

Yes the suggestion is for @bagder's commit message.

I tried ./runtests.pl 100 on Arch linux: it fails leaving log/valgrind100:

valgrind:  Fatal error at startup: a function redirection
valgrind:  which is mandatory for this platform-tool combination
valgrind:  cannot be set up.  Details of the redirection are:
valgrind:  
valgrind:  A must-be-redirected function
valgrind:  whose name matches the pattern:      strlen
valgrind:  in an object with soname matching:   ld-linux-x86-64.so.2
valgrind:  was not found whilst processing
valgrind:  symbols from the object with soname: ld-linux-x86-64.so.2
valgrind:  
valgrind:  Possible fixes: (1, short term): install glibc's debuginfo
valgrind:  package on this machine.  (2, longer term): ask the packagers
valgrind:  for your Linux distribution to please in future ship a non-
valgrind:  stripped ld.so (or whatever the dynamic linker .so is called)
valgrind:  that exports the above-named function using the standard
valgrind:  calling conventions for this platform.  The package you need
valgrind:  to install for fix (1) is called
valgrind:  
valgrind:    On Debian, Ubuntu:                 libc6-dbg
valgrind:    On SuSE, openSuSE, Fedora, RHEL:   glibc-debuginfo
valgrind:  
valgrind:  Note that if you are debugging a 32 bit process on a
valgrind:  64 bit system, you will need a corresponding 32 bit debuginfo
valgrind:  package (e.g. libc6-dbg:i386).
valgrind:  
valgrind:  Cannot continue -- exiting now.  Sorry.

IMO this is an arch+valgrind specific problem. It should not happen. The comment says the debuginfod solution is a short term one, but something else should be done for the long term. Maybe a bug report to arch linux maintainers would be appropriate ?

In addition, it reports searching strlen in ld-linux, which sounds a bit odd unless the dynamic linker holds the other libraries' symbol tables. I checked libc.so.6 and it does export strlen.

I also tried to run DEBUGINFOD_URLS= valgrind /bin/bash from a console and it gives me the same message.

Are there (many) users who are likely to have DEBUGINFOD_URLS set and thus will suffer performance loss if we do this change? If so, why is that variable set if they don't want it used?

I don't understand why Arch wants this weird and special behavior, but I prefer not to make a judgement.

I don't understand why Arch wants this weird and special behavior

I'm not used to this distro, but I think they don't "want" it and it's just a bug.
Package valgrind can be installed without dependency to debuginfod, but does not work without the latter.

If so, why is that variable set if they don't want it used?

That variable is not set by default.
https://wiki.archlinux.org/title/Debuginfod

You have to explicitly install debuginfod to get that environment variable.

I don't understand why Arch wants this weird and special behavior, but I prefer not to make a judgement.

Just a general thing on debuginfod as i'm diving into that to figure out it's benefits. I stumbled upon this blog post:
https://developers.redhat.com/blog/2019/10/14/introducing-debuginfod-the-elfutils-debuginfo-server

From which i want to highlight this part specifically:

For example, your distro might package debuginfo and source files separately from the executable you're trying to debug and you may lack the permissions to install these packages. Or, perhaps you're debugging within a container that was not built with these resources, or maybe you simply don't want these files taking up space on your machine.

Debuginfo files are notorious for taking up large amounts of space, and it is not unusual for their size to be five to fifteen times that of the corresponding executable. debuginfod aims to resolve these problems.

It looks like Arch (and the distros using debuginfod) are ahead of the curve here causing some growing pains. The problems debuginfod tries to solve (according to the above quote) are great to have solved.

If performance is an issue then i'd suggest reaching out to the developers of debuginfod and voice your concerns there. They are probably happy to hear real world usecases.

I don't think the debuginfod is slow. But downloading potentially hundreds of megabytes of debug symbols may be a problem on slow connections. I'd rather install the required symbols given a choice.
That's not to say that the debuginfod is not useful. It solves some problems while creating others and the trick is to find a golden middle.

Thanks. All feedback combined, I believe this is the right PR to merge. I've amended the commit message with @jay's feedback.

For those who are interested in the Arch Linux problem:

• There's already a bug report for it since May 15th: https://bbs.archlinux.org/viewtopic.php?id=276422
• The culprit is commit archlinux/svntogit-packages@b2d3939

Let's monitor how this PR fares and see if we need to take further actions down the road...