- Pharo-dev - lists.pharo.org

Re: [Pharo-project] [ANN]

by Igor Stasenko

2009/9/3 Michael van der Gulik <mikevdg(a)gmail.com>: > Argh. Please reply below other people's emails. > > >> Bill >> >> ________________________________ >> From: pharo-project-bounces(a)lists.gforge.inria.fr >> [mailto:pharo-project-bounces@lists.gforge.inria.fr] On Behalf Of Michael >> van der Gulik >> Sent: Thursday, September 03, 2009 5:54 PM >> To: Pharo-project(a)lists.gforge.inria.fr >> Subject: Re: [Pharo-project] [ANN] >> >> >> >> 2009/9/4 Miguel Enrique CobÃ¡ Martinez <miguel.coba(a)gmail.com> >>> >>> El vie, 04-09-2009 a las 09:33 +1200, Michael van der Gulik escribiÃ³: >>> > >>> > >>> > On Fri, Sep 4, 2009 at 3:21 AM, StÃ©phane Ducasse >>> > <stephane.ducasse(a)inria.fr> wrote: >>> > Â Â Â Â - Issue 1144: Â Many occurrences of "ReadStream on: >>> > Â Â Â Â aCollection" which >>> > Â Â Â Â can be replaced with "aCollection readStream" >>> > >>> > >>> > For what it's worth, I prefer the style of "ReadStream on: xxx". I >>> > prefer to choose the stream class rather than let the collection >>> > choose it for me. >>> >>> But don't you always have the option to call ReadStream on: object, >>> even though object class has a readStream method? >> >> >> In Pharo, sure. >> >> However, in SecureSqueak, there won't be >>readStream methods. I'll spare >> the details because I hate multi-page emails. Ask me if you're interested. > > >> >> 2009/9/4 Schwab,Wilhelm K <bschwab(a)anest.ufl.edu> >> Ok, I'm interested :)Â Is there a security argument?Â Off the top, it does >> not sound like something I would want to use.Â If you can make a case that >> #readStream is dangerous, I'm all ears.Â If it is simply that it closes off >> modifications to the base classes, it is probably too restictive for my >> tastes.Â Locking it down **after** I add the missing methods and deploy on >> a server, that's another matter.Â But then, there would be #readStream >> methods... > > > Excuse the funny formatting. GMail sometimes doesn't cooperate. Also, this > is getting a bit off topic on the Pharo list. > > No, it's not really to do with security. It's to do with this package system > I invented. > > Basically, I want to reduce package dependencies. Putting >>readStream in > kernel classes would make a package dependency between the Collection > package and the Stream package. If the Collections package was loaded into > SecureSqueak, the package architecture would be forced to load the Stream > package as well even if it isn't used. If your own package depended on both > Collection and a different Streams package, anybody who loaded your code > would need to load two Streams packages. > > There's no reason that >>readStream couldn't be included in a Collections > package, provided that it's not in the Collections package provided with the > SecureSqueak Kernel. I want the Kernel package to have as few dependencies > as possible. Packages are read-only in SecureSqueak and the Kernel package > contains special objects, meaning that Kernel and its dependencies can not > be upgraded without distributing a new image. Collections would be a > non-upgradable dependency of Kernel, so I'd be trying hard to not also make > Streams also a non-upgradable dependency. > Interesting dilemma. Right thing, that the package who responsible for implementing #readStream should be the one, who defines any streams of sort. So, then collection could delegate to specific class the responsibility for returning a stream. Following the same logic, an Object class should not contain things like #isCollection, if we imagine things w/o collections, and then load a Collections package, it should provide this method by extending existing behavior. But the problem is, that you need to extend an existing behavior, and once you'll allow that, you are not in secure shell anymore. I think there is solution. By example: The Collection class , in Collection class could define a class var, like ReadStreamDelegate and then a simple method which allows to assign object to it. And then, #readStream could be implemented as: readStream ^ ReadStreamDelegate on: self just don't say that classvars is read-only as well :) You still have to make something non-readonly. So, if classvar is readonly, then i could use one of a slots of an object held in classvar to keep value there :) > More info: > http://gulik.pbwiki.com/Namespaces > http://gulik.pbwiki.com/Packages > > Gulik. > > -- > http://gulik.pbwiki.com/ > > _______________________________________________ > Pharo-project mailing list > Pharo-project(a)lists.gforge.inria.fr > http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project > -- Best regards, Igor Stasenko AKA sig.

Sept. 4, 2009

Re: [Pharo-project] [ANN]

by Peter Hugosson-Miller

2009/9/4 Michael van der Gulik <mikevdg(a)gmail.com> Yes, my answer was meant for you. > > 2009/9/4 Peter Hugosson-Miller <oldmanlink(a)gmail.com> > >> >> Michael, >> > > Assuming you're referring to me. There're lots of Michaels around. > > >> >> This is one of those cases that really makes me miss ENVY, and its Class >> extentions. Under that system you would avoid the problem completely, by >> extending the Collection class in the Stream package and adding the methods >> that depend on Streams in that package. >> >> I guess everyone knows what ENVY Class extensions are, but just in case >> someone doesn't, they are basically collections of methods to be added to a >> Class already in the image (i.e. not subclasses), when the extension is >> loaded. >> > > > Unless there's more magic here than I understand, Monticello also does > this. > I'm *really* new to PharO and Squeak, Monticello and lots of other things, so I don't feel qualified to say one way or the other. > > My package system can't do that. I can't work out a way to modify existing > classes that won't cause complete chaos. SecureSqueak is designed to run > completely untrusted code loaded from the Internet; allowing loaded code to > override methods in other packages allows for too much bad stuff to happen. > It isn't a question of overriding specifically, but of adding methods to classes (which may then of course override methods in their superclasses). But since I don't know enough about SecureSqueak either, I don't really understand what your packaging system is intending to do either. But now I'm afraid I've really dragged the discussion off-topic, for which I apologise :-) > > If you can think of a way to do that securely, I'm all ears. > > > Gulik. > > -- > http://gulik.pbwiki.com/ > > _______________________________________________ > Pharo-project mailing list > Pharo-project(a)lists.gforge.inria.fr > http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project > -- Cheers, Peter

Sept. 4, 2009

[Pharo-project] Added package KernelTests-PeterHugossonMiller.145.mcz

by Peter Hugosson-Miller

Name: KernelTests-PeterHugossonMiller.145 Author: PeterHugossonMiller Time: 4 September 2009, 2:11:41 am UUID: 1bab2752-fa51-48ab-a64d-569f6ceda5bd Ancestors: KernelTests-PeterHugossonMiller.144 Fix for Issue 1150 - Incorrect test that was meant to test Duration>>#storeOn: actually tested #printOn: instead -- Cheers, Peter

Sept. 4, 2009

Re: [Pharo-project] [ANN]

by Michael van der Gulik

2009/9/4 Peter Hugosson-Miller <oldmanlink(a)gmail.com> > > Michael, > Assuming you're referring to me. There're lots of Michaels around. > > This is one of those cases that really makes me miss ENVY, and its Class > extentions. Under that system you would avoid the problem completely, by > extending the Collection class in the Stream package and adding the methods > that depend on Streams in that package. > > I guess everyone knows what ENVY Class extensions are, but just in case > someone doesn't, they are basically collections of methods to be added to a > Class already in the image (i.e. not subclasses), when the extension is > loaded. > Unless there's more magic here than I understand, Monticello also does this. My package system can't do that. I can't work out a way to modify existing classes that won't cause complete chaos. SecureSqueak is designed to run completely untrusted code loaded from the Internet; allowing loaded code to override methods in other packages allows for too much bad stuff to happen. If you can think of a way to do that securely, I'm all ears. Gulik. -- http://gulik.pbwiki.com/

Sept. 4, 2009

Re: [Pharo-project] [ANN]

by Peter Hugosson-Miller

2009/9/4 Michael van der Gulik <mikevdg(a)gmail.com> > Argh. Please reply below other people's emails. > > > Bill >> >> >> >> ------------------------------ >> *From:* pharo-project-bounces(a)lists.gforge.inria.fr [mailto: >> pharo-project-bounces(a)lists.gforge.inria.fr] *On Behalf Of *Michael van >> der Gulik >> *Sent:* Thursday, September 03, 2009 5:54 PM >> *To:* Pharo-project(a)lists.gforge.inria.fr >> *Subject:* Re: [Pharo-project] [ANN] >> >> >> >> 2009/9/4 Miguel Enrique CobÃ¡ Martinez <miguel.coba(a)gmail.com> >> >>> El vie, 04-09-2009 a las 09:33 +1200, Michael van der Gulik escribiÃ³: >>> > >>> > >>> > On Fri, Sep 4, 2009 at 3:21 AM, StÃ©phane Ducasse >>> > <stephane.ducasse(a)inria.fr> wrote: >>> > - Issue 1144: Many occurrences of "ReadStream on: >>> > aCollection" which >>> > can be replaced with "aCollection readStream" >>> > >>> > >>> > For what it's worth, I prefer the style of "ReadStream on: xxx". I >>> > prefer to choose the stream class rather than let the collection >>> > choose it for me. >>> >>> But don't you always have the option to call ReadStream on: object, >>> even though object class has a readStream method? >>> >> >> >> In Pharo, sure. >> >> However, in SecureSqueak, there won't be >>readStream methods. I'll spare >> the details because I hate multi-page emails. Ask me if you're interested. >> > > >> 2009/9/4 Schwab,Wilhelm K <bschwab(a)anest.ufl.edu> >> Ok, I'm interested :) Is there a security argument? Off the top, it >> does not sound like something I would want to use. If you can make a case >> that #readStream is dangerous, I'm all ears. If it is simply that it closes >> off modifications to the base classes, it is probably too restictive for my >> tastes. Locking it down **after** I add the missing methods and deploy on >> a server, that's another matter. But then, there would be #readStream >> methods... >> >> > Excuse the funny formatting. GMail sometimes doesn't cooperate. Also, this > is getting a bit off topic on the Pharo list. > > No, it's not really to do with security. It's to do with this package > system I invented. > > Basically, I want to reduce package dependencies. Putting >>readStream in > kernel classes would make a package dependency between the Collection > package and the Stream package. If the Collections package was loaded into > SecureSqueak, the package architecture would be forced to load the Stream > package as well even if it isn't used. If your own package depended on both > Collection and a different Streams package, anybody who loaded your code > would need to load two Streams packages. > > There's no reason that >>readStream couldn't be included in a Collections > package, provided that it's not in the Collections package provided with the > SecureSqueak Kernel. I want the Kernel package to have as few dependencies > as possible. Packages are read-only in SecureSqueak and the Kernel package > contains special objects, meaning that Kernel and its dependencies can not > be upgraded without distributing a new image. Collections would be a > non-upgradable dependency of Kernel, so I'd be trying hard to not also make > Streams also a non-upgradable dependency. > > More info: > http://gulik.pbwiki.com/Namespaces > http://gulik.pbwiki.com/Packages > > Gulik. > > -- > http://gulik.pbwiki.com/ > > _______________________________________________ > Pharo-project mailing list > Pharo-project(a)lists.gforge.inria.fr > http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project > Michael, This is one of those cases that really makes me miss ENVY, and its Class extentions. Under that system you would avoid the problem completely, by extending the Collection class in the Stream package and adding the methods that depend on Streams in that package. I guess everyone knows what ENVY Class extensions are, but just in case someone doesn't, they are basically collections of methods to be added to a Class already in the image (i.e. not subclasses), when the extension is loaded. -- Cheers, Peter

Sept. 3, 2009

Re: [Pharo-project] [ANN]

by Michael van der Gulik

Argh. Please reply below other people's emails. Bill > > > > ------------------------------ > *From:* pharo-project-bounces(a)lists.gforge.inria.fr [mailto: > pharo-project-bounces(a)lists.gforge.inria.fr] *On Behalf Of *Michael van > der Gulik > *Sent:* Thursday, September 03, 2009 5:54 PM > *To:* Pharo-project(a)lists.gforge.inria.fr > *Subject:* Re: [Pharo-project] [ANN] > > > > 2009/9/4 Miguel Enrique CobÃ¡ Martinez <miguel.coba(a)gmail.com> > >> El vie, 04-09-2009 a las 09:33 +1200, Michael van der Gulik escribiÃ³: >> > >> > >> > On Fri, Sep 4, 2009 at 3:21 AM, StÃ©phane Ducasse >> > <stephane.ducasse(a)inria.fr> wrote: >> > - Issue 1144: Many occurrences of "ReadStream on: >> > aCollection" which >> > can be replaced with "aCollection readStream" >> > >> > >> > For what it's worth, I prefer the style of "ReadStream on: xxx". I >> > prefer to choose the stream class rather than let the collection >> > choose it for me. >> >> But don't you always have the option to call ReadStream on: object, >> even though object class has a readStream method? >> > > > In Pharo, sure. > > However, in SecureSqueak, there won't be >>readStream methods. I'll spare > the details because I hate multi-page emails. Ask me if you're interested. > > 2009/9/4 Schwab,Wilhelm K <bschwab(a)anest.ufl.edu> > Ok, I'm interested :) Is there a security argument? Off the top, it does > not sound like something I would want to use. If you can make a case that > #readStream is dangerous, I'm all ears. If it is simply that it closes off > modifications to the base classes, it is probably too restictive for my > tastes. Locking it down **after** I add the missing methods and deploy on > a server, that's another matter. But then, there would be #readStream > methods... > > Excuse the funny formatting. GMail sometimes doesn't cooperate. Also, this is getting a bit off topic on the Pharo list. No, it's not really to do with security. It's to do with this package system I invented. Basically, I want to reduce package dependencies. Putting >>readStream in kernel classes would make a package dependency between the Collection package and the Stream package. If the Collections package was loaded into SecureSqueak, the package architecture would be forced to load the Stream package as well even if it isn't used. If your own package depended on both Collection and a different Streams package, anybody who loaded your code would need to load two Streams packages. There's no reason that >>readStream couldn't be included in a Collections package, provided that it's not in the Collections package provided with the SecureSqueak Kernel. I want the Kernel package to have as few dependencies as possible. Packages are read-only in SecureSqueak and the Kernel package contains special objects, meaning that Kernel and its dependencies can not be upgraded without distributing a new image. Collections would be a non-upgradable dependency of Kernel, so I'd be trying hard to not also make Streams also a non-upgradable dependency. More info: http://gulik.pbwiki.com/Namespaces http://gulik.pbwiki.com/Packages Gulik. -- http://gulik.pbwiki.com/

Sept. 3, 2009

Re: [Pharo-project] [ANN]

by Schwab,Wilhelm K

Sig, I doubt you missed anything; I dragged EndOfStreamError into it because IMHO, that's where the real problems wait for us. If I had to make code work on 3.10, I would fill in missing pieces (in this case #readStream) there rather than hold back my Pharo code; in fact, that's more or less what I am doing with Pharo too. Bill -----Original Message----- From: pharo-project-bounces(a)lists.gforge.inria.fr [mailto:pharo-project-bounces@lists.gforge.inria.fr] On Behalf Of Igor Stasenko Sent: Thursday, September 03, 2009 6:42 PM To: Pharo-project(a)lists.gforge.inria.fr Subject: Re: [Pharo-project] [ANN] 2009/9/3 Schwab,Wilhelm K <bschwab(a)anest.ufl.edu>: > Sig, > > One of the great things about Smalltalk is that we are seldom forced to do anything, and certainly not in this case. Â It is easy to define #readStream where appropriate, and IMHO Pharo should do so. Â Not only does it give cleaner code, it will help us insert Nile when the time comes, and we can simply add the missing methods. > I understanding this, but at that moment i weren't sure if this is a missing method, or it is missing by intent. Also i had to make sure that my code works also for 3.10. So, for me it was an easiest thing to leave things as they are. > I submit that the problem is instead all of the external code that will still explicitly reference ReadStream. Â That is not a reason to not make the change, and I am hoping that a little leadership by example will cause the right thing to happen across the board. > Right: i'm trying to not use globals if its possible, except well known classes like Arrays, collections etc. > More challenging still are the inconsistencies in exhaustion behavior. Â Referring to my proposal on the wiki, I stand ready to help if we want to enforce EndOfStreamError on any "unauthorized" exhaustion, but I have my doubts that we will do so; Cincom considered and declined the proposal. Â I can also not afford to wait for the decision to be made and imlemented. Â For now, I am gradually moving my code toward #nextOne, #nextMany:, etc., which again can be done simply by adding new code. Â If Pharo adopts EndOfStreamError, I can easily change from my new protocol to the more generic messages. i missed this part of discussion, so i can't comment the above. > > FWIW, one of the first things I did was define #readStream on various collections and collection classes. Â Dolphin (sorry to keep beating the drum, but it sets a wonderful example!) has had #readStream "everywhere" for a long time if not from its earliest days. > The arguments like 'it is done like that somewhere' is a bad way to convince me (and i assume many others). I think you are already given enough arguments before and they already good enough to convince everyone (i hope), except me, because you don't need to convince me on that :) > Bill -- Best regards, Igor Stasenko AKA sig. _______________________________________________ Pharo-project mailing list Pharo-project(a)lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

Sept. 3, 2009

Re: [Pharo-project] [ANN]

by Igor Stasenko

2009/9/3 Schwab,Wilhelm K <bschwab(a)anest.ufl.edu>: > Sig, > > One of the great things about Smalltalk is that we are seldom forced to do anything, and certainly not in this case. Â It is easy to define #readStream where appropriate, and IMHO Pharo should do so. Â Not only does it give cleaner code, it will help us insert Nile when the time comes, and we can simply add the missing methods. > I understanding this, but at that moment i weren't sure if this is a missing method, or it is missing by intent. Also i had to make sure that my code works also for 3.10. So, for me it was an easiest thing to leave things as they are. > I submit that the problem is instead all of the external code that will still explicitly reference ReadStream. Â That is not a reason to not make the change, and I am hoping that a little leadership by example will cause the right thing to happen across the board. > Right: i'm trying to not use globals if its possible, except well known classes like Arrays, collections etc. > More challenging still are the inconsistencies in exhaustion behavior. Â Referring to my proposal on the wiki, I stand ready to help if we want to enforce EndOfStreamError on any "unauthorized" exhaustion, but I have my doubts that we will do so; Cincom considered and declined the proposal. Â I can also not afford to wait for the decision to be made and imlemented. Â For now, I am gradually moving my code toward #nextOne, #nextMany:, etc., which again can be done simply by adding new code. Â If Pharo adopts EndOfStreamError, I can easily change from my new protocol to the more generic messages. i missed this part of discussion, so i can't comment the above. > > FWIW, one of the first things I did was define #readStream on various collections and collection classes. Â Dolphin (sorry to keep beating the drum, but it sets a wonderful example!) has had #readStream "everywhere" for a long time if not from its earliest days. > The arguments like 'it is done like that somewhere' is a bad way to convince me (and i assume many others). I think you are already given enough arguments before and they already good enough to convince everyone (i hope), except me, because you don't need to convince me on that :) > Bill -- Best regards, Igor Stasenko AKA sig.

Sept. 3, 2009

Re: [Pharo-project] FFI meets 74 bits - there be dragons here??

by Schwab,Wilhelm K

Obviously I mean 64 bits :) -----Original Message----- From: pharo-project-bounces(a)lists.gforge.inria.fr [mailto:pharo-project-bounces@lists.gforge.inria.fr] On Behalf Of Schwab,Wilhelm K Sent: Thursday, September 03, 2009 6:34 PM To: Pharo-project(a)lists.gforge.inria.fr Subject: [Pharo-project] FFI meets 74 bits - there be dragons here?? Hello all, So far, I am one for two on getting ODBC to work on my 64 bit Ubuntu system. I'm using a pre-production (the only option we were given by our IT guys) Cisco vpn that does not support 64 bit systems; enter a 32 bit vpn and some copied libraries. After that fun, I have been able to get isql (unixODBC test utility) to see a MySQL server I have living on a blade machine. That much works. My Pharo image seems unable to talk to it, and I am wondering whether it might be because the library is 64 bit - please excuse me if that's just plain stupid. Of course, so was installing the 64 bit OS<g>, but I figure there will not be an option some day, so I'm trying to get my bad news early. Any ideas about how to proceed? I'm thinking of subclassing ODBCLibrary and tweaking the types (something on the net addresses some changes) to see if I can get some success and then keep going as needed. Does that sound reasonable? Any better ideas? FreeTDS is letting me down talking to a SQL server machine. If anyone has set that up successfully, I would greatly enjoy knowing what you did. Bill _______________________________________________ Pharo-project mailing list Pharo-project(a)lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

Sept. 3, 2009

[Pharo-project] FFI meets 74 bits - there be dragons here??

by Schwab,Wilhelm K

Hello all, So far, I am one for two on getting ODBC to work on my 64 bit Ubuntu system. I'm using a pre-production (the only option we were given by our IT guys) Cisco vpn that does not support 64 bit systems; enter a 32 bit vpn and some copied libraries. After that fun, I have been able to get isql (unixODBC test utility) to see a MySQL server I have living on a blade machine. That much works. My Pharo image seems unable to talk to it, and I am wondering whether it might be because the library is 64 bit - please excuse me if that's just plain stupid. Of course, so was installing the 64 bit OS<g>, but I figure there will not be an option some day, so I'm trying to get my bad news early. Any ideas about how to proceed? I'm thinking of subclassing ODBCLibrary and tweaking the types (something on the net addresses some changes) to see if I can get some success and then keep going as needed. Does that sound reasonable? Any better ideas? FreeTDS is letting me down talking to a SQL server machine. If anyone has set that up successfully, I would greatly enjoy knowing what you did. Bill

Sept. 3, 2009