HN Explorer

Ask HN: Can someone explain why this line exists?

25 points by andrewgross ↗ HN
https://github.com/mongodb/mongo-java-driver/blob/master/src/main/com/mongodb/ConnectionStatus.java#L213

Found via: https://twitter.com/avibryant/status/339840598871769088

16 comments

[ 6.4 ms ] story [ 23.6 ms ] thread
byoung2
If an exception was thrown, and _ok evaluates to false, then approximately 10% of the time, log the exception to the error log. Probably just to cut down on noise in the logs, or to account for brief periods of latency where the server appears down, but isn't.
andrewgross
Thanks, was reviewing it with a coworker and crept up on the same conclusion. Seems a bit hard to test and definitely alarming at first glance.
avibryant
I think that you're right about the intent of the author, but unless I'm mistaken, that's not what the code does. If _ok is false, it boils down to if(!(Math.random() > 0.1))) which is the same as if(Math.random() <= 0.1), which means it will return early only 10% of the time, so the exception gets logged 90% of the time.

Wait, what?

zaroth
Upvote to you, and downvote to whoever wrote that line of code.

A ternary inside an 'if' statement, really? And even more so when one of the return values is 'true'?

Relax, it's just a bug, right? But it would be nice if programmers practiced basic Boolean reduction in their 'if' statements...

(!((_ok) ? true : (Math.random() > 0.1)))

!((_ok) ? true : (Math.random() > 0.1))

_ok ? false : !(Math.random() > 0.1)

!_ok && Math.random() <= 0.1

dampier
If you read the code, you will notice that the FIRST time this error is encountered, it will always be logged.

The test in question is done frequently, and subsequent failures are about 100 times less interesting than the first. Unless you just like filling up disks with log files while your network is acting up.

Finally, note that the "_ok" flag will be RESET and the resolved condition logged once the test finally returns to succeeding.

andrewgross
Awesome, thanks for the explanation
lotyrin
Thank you for an excellent counterexample for when people claim code comments are useless.
dampier
Heh ... agreed. A comment here would definitely not go amiss.
ollysb
You could express the same thing without comments:

    boolean alreadyLogged = !_ok;
    boolean skipLogging = alreadyLogged && (Math.random() > 0.1)
    if( skipLogging ) return res;
Dylan16807
I strongly dislike your version. Creating a boolean just to use and discard in the next line? Twice? Ugly. I'd much rather keep the code simple and try to cut out confusing parts. Give the flag a better name and remove the useless ternary. I'd also get rid of the conditional return and let the one outside the catch do the work.

  if (!errorLogged || (Math.random() < 0.1)) {
    
    errorLogged = true;
    
    // log error
    
  }
But if it's a choice between comments and adding temporary variables that don't do anything, I'll almost always choose comments.
ollysb
I wouldn't actually have them as Boolean variables, I'd extract them into query methods, then the code would just be:

    if( skipLogging() ) return res;
qu4z-2
I'd tend to choose temporary variables. I trust my compiler to optimise them out, and I find it makes it much easier to find the appropriate place to change or add new code.

"Does this affect working out whether the error has been logged before?", "Does this affect whether we should log this error?", etc

EDIT: But in this case I'd still want to put a comment clarifying the "why".

ranman
are you sure the compiler is smart enough to do that? I think we frequently assume compilers are smarter than they really are because we don't fully understand them anymore... but in practice I find they often do very dumb things.
lmm
No, but I am sure that my automated performance testing procedures will detect if I've introduced a significant performance problem, and it would be an inefficient use of my time to worry about insignificant performance effects.
Dylan16807
Register renaming is one of the most very basic levels of optimization. I challenge you to find me a single optimizing compiler that slows down when you give names to temporary results. Hell, I can show you non-optimizing compilers that suffer no slowdown from code like this.
zaroth
I commented earlier, but I can't get over how ridiculous that line is. If you started by just writing the intent (!ok && rand() > .1) you actually have to work really hard to turn it into what you have there, and introduce a bug in the process.

Reminds me of this: http://www.osnews.com/story/19266/WTFs_m - "Dude, WTF!"