random engine #9016
random engine #9016
Conversation
|
This pull request introduces 1 alert when merging c1fc00a into c1fdf91 - view on LGTM.com new alerts:
|
| } | ||
| return *this; | ||
| } | ||
| DNSName(const DNSName& a) = default; |
There was a problem hiding this comment.
I know this is a work-in-progress but be careful that declaring a copy assignment operator or a copy constructor will prevent the move operations from being generated.
There was a problem hiding this comment.
Yes. But so far this is the only "solution" I could think of.
omoerbeek
force-pushed
the
random-engine
branch
from
79e87dc to
c49188f
Compare
There was a problem hiding this comment.
The general concept of this PR looks good to me. A few nits, and also it seems to miss quite a few other users of std::random_shuffle():
- pdns/calidns.cc
- pdns/dnsdist-lua-actions.cc
- pdns/ws-auth.cc
- pdns/signingpipe.cc
- pdns/dynhandler.cc
- pdns/dnsbulktest.cc
- pdns/slavecommunicator.cc
| return *this; | ||
| } | ||
| DNSName(const DNSName& a) = default; | ||
| DNSName(DNSName&& a) = default; |
There was a problem hiding this comment.
| DNSName(DNSName&& a) = default; | |
| DNSName(DNSName&& a) = default; | |
| DNSName& operator=(DNSName&& rhs) | |
| { | |
| if (this != &rhs) { | |
| d_storage = std::move(rhs.d_storage); | |
| } | |
| return *this; | |
| } |
| #include "dnsparser.hh" | ||
|
|
||
| // shuffle, maintaining some semblance of order | ||
| void pdns::shuffle(std::vector<DNSZoneRecord>& rrs) |
There was a problem hiding this comment.
So if I understand correctly, this function assumes that we have first CNAMEs in the ANSWER section, then other ANSWER records? In which case it will then shuffle the second ones only. And Then then the same for additional records? I know it's has not been modified by this PR but I think some comment might help for the next time, especially since the move to a new file means the history will be harder to find.
pdns/shuffle.cc
Outdated
|
|
||
|
|
||
| // shuffle, maintaining some semblance of order | ||
| void pdns::shuffle(std::vector<DNSRecord>& rrs) |
There was a problem hiding this comment.
And this one is mostly the same, except it also assumes that RRSIGs will come at the end of each section. It looks like it's never called directly, always from orderAndShuffle(), perhaps we could make it static?
pdns/dnsdist-lua-actions.cc
Outdated
| @@ -553,8 +554,9 @@ DNSAction::Action SpoofAction::operator()(DNSQuestion* dq, std::string* ruleresu | |||
| } | |||
| } | |||
|
|
|||
| static std::default_random_engine r; | |||
There was a problem hiding this comment.
Unless std::default_random_engine::operator() is thread-safe, I don't think we should make r static.
There was a problem hiding this comment.
Indeed. I do not think we want to re-create the engine object all the time, since it might use a static seed (and it does afaks in at least the clang c++ lib). Maybe a thread local?
There was a problem hiding this comment.
How about placing it in the SpoofAction class? I can write that change if you would like me to :)
There was a problem hiding this comment.
Right. If that takes care of the multi-threaded access that is preferred. Please do so. There might be caveats I do not know with the DNSAction classes :-)
There was a problem hiding this comment.
For a reason unknown to me Github doesn't allow me to open a pull request against your branch, so please pull 387221d from https://github.com/rgacogne/pdns/tree/random-engine :)
|
I think it is good to merge now. I'll squash it first though. |
std::random_shuffle is deprecated and removed starting from C++17, so we might better use std::shuffle which uses URBG: https://en.cppreference.com/w/cpp/named_req/UniformRandomBitGenerator under the hood for a better random generation.
…tring do not like that and throw an assertion.
…ndom_engine for shuffle.
omoerbeek
force-pushed
the
random-engine
branch
from
387221d to
77d574a
Compare
Short description
Continuation of #9004, which does not allow me to push anymore.
Checklist
I have: