On Tue, Apr 12, 2011 at 6:40 PM, Jon Kristensen wrote:

Hello Haskellers!

As I have now implemented everything that I could think of, I would like to ask the Haskell community to give some feedback on HLogger so that I can continuing improving it. Some questions:

Do you have any general/architectural thoughts on things I might be doing wrong?

1) Use "Text" as the string type, not "String" 2) Re: SimpleLogger: writing to and flushing the log file upon receiving each message is absolutely the wrong thing to do in high-performance code. Each write() is a context switch into the kernel, and doing it this way will kill your throughput (in messages per second). What we do in our webserver (which relies on high-throughput access logging) is to buffer messages to the log channel in memory, and perform the actual writes in a separate worker thread no more often than once every N seconds.

Is the ease-of-use goal contradicting making this library useful in more complex applications?

"Easy to use" and "high performance" don't have to be contradictory, but you shouldn't neglect the latter. G -- Gregory Collins

On Tue, Apr 12, 2011 at 11:40 AM, Jon Kristensen wrote:

Hello Haskellers!

I have developed a very simple logging library to be extendable and easy-to-use. You can see the current release at http://hackage.haskell.org/package/hlogger/. The SimpleHLogger module is an example of what a HLogger implementation could look like. It could be used like this:

import System.Log.SimpleHLogger

main = do

  logger <- simpleLogger "test"

  loggerLog logger (Just ("LoggerTest", "main")) Debug "Test"

  loggerStop logger

As I have now implemented everything that I could think of, I would like to ask the Haskell community to give some feedback on HLogger so that I can continuing improving it. Some questions:

It looks very simple, which is nice. Just about the only things I would add: * A standard-out/standard-error logger * A monoid instance for the Logger type * Convinience functions for logging without context and logging at a specific error level:

logError myLogger "Oh no!" logDebug myLogger "a thing happened"

These might be in a separate module, so I can define them for myslef based on some logContext function, so that in my module I can define:

logError msg = logContext "Module.Name" LogLevelError msg

You might even be able to offer a separate package with TemplateHaskell splices to build these helpers. Maybe I'm going off the deep-end it terms of complexity. Have you looked into HsLogger at all? This project has a confusingly similar name - so much so I thought that your email was about HsLogger! Take care, Antoine

Do you have any general/architectural thoughts on things I might be doing wrong? Is the ease-of-use goal contradicting making this library useful in more complex applications? Do I want to redesign HLogger to allow loggers to have a mutable state, and if so, how can I do that? Is there a nice way to provide the module and the function than using (Just ("LoggerTest, "main)) (shortening it somehow, I guess)?

Thanks!

Warm regards, Jon Kristensen

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe