David and Keith: Thanks you for your reply.

I will see and try to use RIO. However, what I need to do is very simple. I have already add log to SqueakDBX. Till now, the log just print in Transcript. The problem is that sometimes I get a segmentation fault and I cannot see transcript any more haha. So, I would like to be able to log into a log file.

If I think I could give some useful feedback be sure I will let you know.

cheers,

Mariano

On Mon, Nov 24, 2008 at 4:53 PM, Keith Hodges keith_hodges@yahoo.co.ukwrote:

Mariano Martinez Peck wrote:

...

Hi list!

I need to do some things with files (text and binaries), and I would need information about creating, deleting, adding, appending, writting and so on with them. I look in SBE but I didn't find a chapter about

that.

...

Is somewhere documentation about this?

thanks,

Mariano

The built in File handling is not very elegent, enter, "Rio" File/Drectory

Installer squeaksource project: 'Rio'; install: 'File-Kernel'; install: 'File-Base'.

(and File-Tests for examples)

see also http://wiki.squeak.org/squeak/5929

if you like it, then perhaps you would be kind enough to offer some feedback

best regards

Keith

========= Class comment from File reads:

Features of File/Directory (née Rio)

File/Directory was inspired by ruby's rio which enables local and remote files to be used transparently. To achieve this a generic stream copy function has been developed (aStream copyTo: bStream) and we are able to expose the working stream of FtpClient so you can stream directly to a remote file. We also support file archives (zip) as if they were a filesystem, and transparent streaming to compressed files (gzip). [transparent streaming to a remote compressed file is still available as a exercise for the reader to implement ;) ]

File/Directory is a standalone package, it does not use any of the existing FileDirectory classes. It is separated into File-Kernel and File-Base, where the former provides a minimal working subset for KernelImages to access local filesystems.

Local file copying is handled by the fileCopyPlugin if it is present, if not OSProcess 'cp' is used, and as a last resort a full stream to stream copyTo: is performed (but this does update modification time).

Remote file upload or download is handled by OSProcess 'ftp' if it is present. Multiple file transfers are invoked N at a time (where N is initially 5). Copying from a remote file to another remote file is handled by the internal stream copy function.

Keith: hahahah. I know there was a "Logging" project, but as there were few places where to log, I decided to do it by hand. Do you think it should be better to do it with it altought I have few places to log ?

Just by curious, is Logging something like log4j (sorry, but that's what I know :( ?

does Logging use RIO for files ?

thanks,

Mariano

On Mon, Nov 24, 2008 at 6:05 PM, Keith Hodges keith_hodges@yahoo.co.ukwrote:

Mariano Martinez Peck wrote:

...

David and Keith: Thanks you for your reply.

I will see and try to use RIO. However, what I need to do is very simple. I have already add log to SqueakDBX. Till now, the log just print in Transcript. The problem is that sometimes I get a segmentation fault and I cannot see transcript any more haha. So, I would like to be able to log into a log file.

If I think I could give some useful feedback be sure I will let you know.

cheers,

Mariano

Ah! Logging! Why didnt you say so.

We would like to move Squeak to use some form of proper logging interface, rather than blindly throwing stuff at transcript show. (which is not thread safe)

The current proposal is "Logging". (Installer install: 'Logging') This provides a front end to the users choice of available logging frameworks. Toothpick, and SimpleLog are currently supported.

Documentation: http://installer.pbwiki.com/Logging

regards

Keith

ok. If you say so, I will try it as soon as I have time and I will let you know my feedback and any question I have with it.

thanks!

mariano

On Mon, Nov 24, 2008 at 6:42 PM, Keith Hodges keith_hodges@yahoo.co.ukwrote:

Mariano Martinez Peck wrote:

...

Keith: hahahah. I know there was a "Logging" project, but as there were few places where to log, I decided to do it by hand. Do you think it should be better to do it with it altought I have few places to log ?

Definitely.

That is exactly what Logging is there for, so that people are encouraged to put the log code in their packages without worrying about how the actual logging is done, timestamped etc. You can let the user choose how the output is provided/used/managed (e.g. SimpleLog does log file rotation). The default backend is simply a Null instance (it should/could be a minimal Transcript Log equivalent to Transcript show:)

There are also utilities for timing events, so you could start a timer when you send a query and print it when you get the result.

...

Just by curious, is Logging something like log4j (sorry, but that's what I know :( ?

does Logging use RIO for files ?

thanks,

Mariano

If you load "SimpleLog" (from the gjallar team) this uses Syslog, transcript, LogMorph or files (not rio, just a basic filestream)

Keith

Göran: Yesterday night we have a very good discussion about logging in IRC. I was asking about Logging in order to add log to SqueakDBX. I think there were some interesting questions:

1) Should a framework, like SqueakDBX, log? Or this is a responsibility of the "user" of the framework ?

Here, I will say: I think framework must log because it helps the framework team to find out bugs or give them information about it. I think this is something good. Obviously the framework should only log things related to the framework itself and not logic related with the user of it.

2) If so, should it be optional ? I think i must. I mean, user can enable or disable it.

3) Suppose I want very much performance, and the user disable logging. Which is the best way of doing this?

4) No matter 1) to 3), framework's user can log. I mean, it can do something like [xxxx] on: DBXError do:[ self log 'something' ]

In fact, I decided (actually I must discuss with the team) to use Logging and SimpleLog in SqueakDBX.

Any opinion about this?

On Tue, Nov 25, 2008 at 8:21 AM, Göran Krampe goran@krampe.se wrote:

Hi!

Mariano Martinez Peck wrote:

...

ok. If you say so, I will try it as soon as I have time and I will let you know my feedback and any question I have with it.

I recommend (of course) using "SimpleLog", and optionally with the "Logging" frontend from Keith.

SimpleLog was developed in the Gjallar project and it has the following goodness:

• Simple. :)
• Uses the standard syslog levels.
• Has a UDP based syslog backend. Fast and very useful.
• Has a Morphic viewer which is quite nice and has filtering abilities.
• Has log file rotation.
• Can still log to Transcript but it is protected by a mutex.

If you need to use external monitoring/alarm tools and/or suddenly need to deploy an app "clustered" then the syslog backend is very nice to have. Otherwise you end up with one log file per "clustered image" which is not so nice.

If you use the "Logging" frontend (which I haven't used myself) then you of course have options to choose backends.

regards, Göran

Hi again!

Mariano Martinez Peck wrote:

Göran: Yesterday night we have a very good discussion about logging in IRC. I was asking about Logging in order to add log to SqueakDBX. I think there were some interesting questions:

1. Should a framework, like SqueakDBX, log? Or this is a responsibility

of the "user" of the framework ?

I think it definitely should log. :)

Here, I will say: I think framework must log because it helps the framework team to find out bugs or give them information about it. I think this is something good. Obviously the framework should only log things related to the framework itself and not logic related with the user of it.

Right.

2. If so, should it be optional ?

I think i must. I mean, user can enable or disable it.

Yep.

3. Suppose I want very much performance, and the user disable logging.

Which is the best way of doing this?

Good question. :) As Keith noted you can do tricks with subclasses etc. I haven't bothered looking into disabling logging "the fastest way" much for SimpleLog, and now since Keith has taken the "front seat" :) perhaps that is now an issue more suitable to deal with in his "Logging" package.

4. No matter 1) to 3), framework's user can log. I mean, it can do

something like [xxxx] on: DBXError do:[ self log 'something' ]

Right.

In fact, I decided (actually I must discuss with the team) to use Logging and SimpleLog in SqueakDBX.

Any opinion about this?

Just happy to hear it! :)

regards, Göran

On Thu, Nov 27, 2008 at 9:44 AM, Keith Hodges keith_hodges@yahoo.co.uk wrote:

self send a message to me. A := self nextThing.

What naughty superpower code can I put in to #send that will jump execution to self nextThing directly?

Have #send return an object which has #doesNotUnderstand: implemented as ^ self.

This is a "message eating nil" - see also:

http://web.archive.org/web/20060223010948/http://www.smalltalkpro.com/genera...

Avi Bryant wrote:

On Thu, Nov 27, 2008 at 9:44 AM, Keith Hodges keith_hodges@yahoo.co.uk wrote:

...

self send a message to me. A := self nextThing.

What naughty superpower code can I put in to #send that will jump execution to self nextThing directly?

Have #send return an object which has #doesNotUnderstand: implemented as ^ self.

This is a "message eating nil" - see also:

Thanks Avi,

Indeed, I am a champion of the message eating null, see http://www.squeaksource.com/Null for squeak.

That's what I am doing already. The problem is that the arguments to these methods may take a long time to calculate so I am looking for a short cut, all beit an evil one.

regards

Keith

On Fri, Nov 28, 2008 at 8:52 AM, Bert Freudenberg bert@freudenbergs.dewrote:

On 27.11.2008, at 18:44, Keith Hodges wrote:

Hello All,

...

I need a villainous superpower that I haven't needed before, can anyone tell me how to achieve this.

I have some code like this:

self send a message to me. self nextThing.

or

self send a message to me. A := self nextThing.

What naughty superpower code can I put in to #send that will jump execution to self nextThing directly?

send ... thisContext sender jump: 4

This skips 4 bytecodes in the calling method before returning, so instead of pc=31 it continues at pc=35:

In terms of code smells, this is chemical warfare!! :-D.

Gulik.

-- http://people.squeakfoundation.org/person/mikevdg http://gulik.pbwiki.com/

...

In terms of code smells, this is chemical warfare!! :-D.

I must admit I am having second thoughts!

No doubt about this. Kids, don't try this at home.

It still tremendous fun that you can do it. Keith, I have no idea what you need this for but I think the following is pretty complete - it skips until the next pop bytecode which marks the end of the current statement. It should get confused only if there was a block in that line that had a pop in itself:

send Transcript show: 'send '. self returnToPop. self neverReached

returnToPop | ctxt | ctxt := thisContext sender sender. [ctxt nextByte = 135] whileFalse: [ctxt jump: 1]. ctxt resume

Try filing in attachment and eval "Evil new test"

• Bert -

Thats really cool, thanks! It gives me a warm fuzzy feeling just to know we can do it.

In the Logging framework, the message eating Null WAS used to disable logging features.

so for example the call...

self log info mysql dumper: theDataReturned asSuitableFormat.

A) to disable the #info level log messages -> MyLogRouterPolicy-#info ^null B) to disable the #mysql category log messages -> MyLogRouterPolicy-#mysql ^null C) to disable all Logging, and even remove the Logging framework completely. Object-#log ^null

Realizing that some user code may take a lot of time, #asSuitableFormat in this example, I began looking for alternative solutions. (hence the question)

Instead of using null as a hammer, I implemented an explicit #isIgnore flag which covers most use cases. (It still cannot cover the case where no Logging framework is loaded at all)

My solution is as follows:

self log info mysql use: [

self log dumper: theDataReturned asSuitableFormat

].

I have a personal convention #use: is my version of #in: which is a bit more flexible, I use it like this e.g.

To ensure streams close:

FileStream-#use: ^ [ aBlock value: self ] ensure: [ self close ].

To provide psuedo null behaviour (equivalent of ifNotNilDo:)

UndefinedObject-#use: aBlock ^ self

So...

LogRouter-#use: aBlock

isIgnore ifFalse: [ aBlock value ]

This allows a user to disable mysql Logging by implementing

MyLogRouterPolicy-#mysql ^ self ignore

With this solution I am almost happy. It is cleaner, and allows whole chunks of logging code to be skipped if the user simply sets part of their LoggingPolicy to self ignore.

So, we can now spit chunks of verbose logging code without being afraid of slowing things down for someone else, since we know that any user is in complete control via their chosen LogRouterPolicy

However, after this I dont think I can resist releasing "SuperNull" sometime soon.

thanks for an interesting insight,

Keith

Bert Freudenberg a écrit :

On 27.11.2008, at 18:44, Keith Hodges wrote:

...

Hello All,

I need a villainous superpower that I haven't needed before, can anyone tell me how to achieve this.

I have some code like this:

self send a message to me. self nextThing.

or

self send a message to me. A := self nextThing.

What naughty superpower code can I put in to #send that will jump execution to self nextThing directly?

send ... thisContext sender jump: 4

This skips 4 bytecodes in the calling method before returning, so instead of pc=31 it continues at pc=35:

29 <70> self 30 <D4> send: send 31 <D3> send: a 32 <D2> send: message 33 <D1> send: to 34 <D0> send: me 35 <87> pop 36 <70> self 37 <D5> send: nextThing 38 <87> pop 39 <78> returnSelf

• Bert -

Maybe something like a classical doesNotUnderstand: handling... A SilentMessageAbsorber could return self if it doesNotUnderstand: a message, so implement send ^SilentMessageAbsorber new

That's not exactly direct, rather nop...

Nicolas

"Keith Hodges" keith_hodges@yahoo.co.uk wrote in message

The built in File handling is not very elegent, enter, "Rio" File/Drectory

Installer squeaksource project: 'Rio'; install: 'File-Kernel'; install: 'File-Base'.

On Damien's 3.10 dev image I get warnings about FileLocalExecutive, which appears undefined both at start and end of the install.

- Sophie