reduce floodding debug output

[GitMensch]

suppress debug output for invalid address passed to memoryRead if it is identical to the last recognized invalid one

[GitMensch]

casing fixed and moved to UnwindInfo as requested; note: this obviously adds much more data (one for each UnwindInfo - but I don't know how often / much of those we allocate) and is "global"; I don't know if the original "shows less often the same entry" is now better (because we get back to the same entry later and already marked it as invalid) or not as good as the original "only cache the very last entry, so you don't get the same output over and over".
(note to self: nice for testing once this lands is named "DOKU_AKS")

[milianw]

casing fixed and moved to UnwindInfo as requested; note: this obviously adds much more data (one for each UnwindInfo - but I don't know how often / much of those we allocate)

We have one of these per process, this is really a tiny amount of data.

and is "global"; I don't know if the original "shows less often the same entry" is now better (because we get back to the same entry later and already marked it as invalid) or not as good as the original "only cache the very last entry, so you don't get the same output over and over". (note to self: nice for testing once this lands is named "DOKU_AKS")

I don't quite know what you are saying here. It sounds like you didn't actually test whether this works as desired? Can you please do that then?

[GitMensch]

For the last test I'll create an "update perfparser" PR to Hotspot after doing the requested changes, if I can't get the binary created in my dev environment working on the machine where I have problems (that only uses the appimages).

In any case:
Could you add some minimal CI here that checks that everything compiles and passes quality checks (copy over from Hotspot) and possibly make the perfparser binary available that "matches" the appimage environment - or would this make upstream work hard?

[GitMensch]

I've added the missing init (I think) and then tested the resulting hotspot-perfparser by replacing the one found in the appimage with this one, strangely instead of

Invalid memory read requested by dwfl fffffffffffffff9
Invalid memory read requested by dwfl fffffffffffffff9
Invalid memory read requested by dwfl fffffffffffffff9
Invalid memory read requested by dwfl fffffffffffffff9
Invalid memory read requested by dwfl fffffffffffffff9
Invalid memory read requested by dwfl fffffffffffffff9
Invalid memory read requested by dwfl fffffffffffffff9
Invalid memory read requested by dwfl fffffffffffffff9
Invalid memory read requested by dwfl fffffffffffffff9
Invalid memory read requested by dwfl fffffffffffffff9

I have zero occurrences of that, while I've expected one.

[GitMensch]

Hm, not sure if there's something broken with this PR or with my build, but using an existing appimage (extracted) and copy-paste a hotspot-perfparser from this build breaks the flamegraph. That needs to be checked before a possible pull.

[GitMensch]

I see no reason why that should break hotspot's flamegraph and I've checked that it does produce less output, so... please review/pull.

[GitMensch]

Still get a bunch of those outputs with Hotspot ;-)
Friendly ping @milianw for re-review, the requested changes were done two weeks ago.

[milianw]

so, code wise this looks good. but:

• you said it breaks stuff? is that really the case? have you investigated that?
• and are you saying this isn't enough and you still get "a bunch of those outputs" with this patch, meaning it's not enough?

[GitMensch]

I had broken results, but this likely comes from my Frankenbuild, I don't see how those coffee changes would be the reason for that. I did see much less warnings with using that (one per different address), but that's expected.