Problem due to deprecation of class Url in Pharo 3 and Moose 5.0

Hello

 

I have run into a problem in moving some existing work from earlier versions of Pharo/Moose. I have found a work around, but I wonder if there is a tidier way of handling it.

 

I make frequent use of Todd Blanchard’s HTML parser and validator, HTMCSS (http://smalltalkhub.com/#!/~ToddBlanchard/HTMCSSValidatingParser), which was originally written for Squeak but has performed without trouble on earlier versions of Pharo. When I try to use it on Moose 5.0, I get frequent warning messages about Url being deprecated. I have made the problem go away by commenting out the deprecation warning in Url class>>new, but I wonder what should be done on a more permanent basis.

Yes change the code of the Parser not to use old code.

The problem arises because HTMCSS does not just parse the original HTML file; it also loads and parses any referenced CSS files. (This is a function I could do without, but I don’t fancy trying surgery on a complex package where I only partly understand the workings!) It constructs the full address of the CSS file by combining the root address of the HTML with the relative address of the CSS, using Url class>>combine:withRelative:, and in the course of this it invokes Url class>>new; hence the deprecation message.

use ZnURL

The deprecation message says Url has been replaced by ZnUrl, but this is clearly  not a simple replacement of one message by an equivalent; there is no ZnUrl class>>combine:withRelative:, for instance.

Sven will certainly comment but I guess that there is certainly the same behavior.

The tidiest solution would no doubt be to find an equivalent method in Zinc, which I am sure exists, and then modify the HTMCSS code to use it. I have tried to find an equivalent, but Zinc is a large and complex system and I rapidly got lost.

It should be in ZnURL

is it not addPathSegment:?


from class comment

  ZnUrl new
    scheme: #https;
    host: 'encrypted.google.com';
    addPathSegment: 'search';
    queryAt: 'q' put: 'Smalltalk';
    yourself.
 
host: looks like the root
addPathSegment: looks like https://encrypted.google.com/search

 

I wonder more broadly about the strategy of deprecating functions which are required by legacy packages, as in this case. Should there at least be a way of overriding deprecations? (I suppose that is what I have done by commenting it out, but it seems crude.)

 

Thanks in advance for any help or suggestions.

 

Peter Kenny

ZnUrl>>#inContextOf: is the selector you are looking for, but path merging is not supported (yet).

'readme.txt' asZnUrl inContextOf: '<a href="http://www.host.com:8080'">http://www.host.com:8080' asZnUrl.

Maybe we should add path merging, I'll think about it.

This is how you can work around Deprecation warnings:

[ Url combine: 'http://www.foo.com/one/two/' withRelative: 'bar/readme.txt' ]
  on: Deprecation do: [ :exception | exception resume ]

but that is a temporary hack.

>> Thanks in advance for any help or suggestions.
>>  
>> Peter Kenny
>

Sven

Sven: Thanks for the suggestions. I have tried ZnUrl>>#inContextOf:, but it
does not produce the correct address; evidently I shall have to wait until
you have full path merging implemented. Meanwhile, I have used your
'temporary hack' to ignore the deprecation; it is at least not as hackish as
my approach.
Stepharo:
> Yes change the code of the Parser not to use old code.
As I pointed out, the Parser is a large and complex package, and I am not
the author, so this is easier said than done. It seems a bit odd to dismiss
code which worked OK in Moose 4.9, but not in Moose 5.0, as 'old code.'  It
is code which has no functional equivalent in the new code, as Sven has
confirmed.

Thanks again

Peter Kenny

-----Original Message-----
From: Pharo-users [mailto:[hidden email]] On Behalf Of
Sven Van Caekenberghe
Sent: 10 January 2015 16:36
To: Any question about pharo is welcome
Subject: Re: [Pharo-users] Problem due to deprecation of class Url in Pharo
3 and Moose 5.0


> On 10 Jan 2015, at 16:45, stepharo <[hidden email]> wrote:
>
>
> Le 10/1/15 11:41, PBKResearch a écrit :
>> Hello
>>  
>> I have run into a problem in moving some existing work from earlier
versions of Pharo/Moose. I have found a work around, but I wonder if there
is a tidier way of handling it.
>>  
>> I make frequent use of Todd Blanchard’s HTML parser and validator, HTMCSS
(http://smalltalkhub.com/#!/~ToddBlanchard/HTMCSSValidatingParser), which
was originally written for Squeak but has performed without trouble on
earlier versions of Pharo. When I try to use it on Moose 5.0, I get frequent
warning messages about Url being deprecated. I have made the problem go away
by commenting out the deprecation warning in Url class>>new, but I wonder
what should be done on a more permanent basis.
> Yes change the code of the Parser not to use old code.
>>
>> The problem arises because HTMCSS does not just parse the original HTML
file; it also loads and parses any referenced CSS files. (This is a function
I could do without, but I don’t fancy trying surgery on a complex package
where I only partly understand the workings!) It constructs the full address
of the CSS file by combining the root address of the HTML with the relative
address of the CSS, using Url class>>combine:withRelative:, and in the
course of this it invokes Url class>>new; hence the deprecation message.
>
> use ZnURL
>>
>> The deprecation message says Url has been replaced by ZnUrl, but this is
clearly  not a simple replacement of one message by an equivalent; there is
no ZnUrl class>>combine:withRelative:, for instance.
> Sven will certainly comment but I guess that there is certainly the same
behavior.

ZnUrl>>#inContextOf: is the selector you are looking for, but path merging
is not supported (yet).

'readme.txt' asZnUrl inContextOf: '<a href="http://www.host.com:8080'">http://www.host.com:8080' asZnUrl.

Maybe we should add path merging, I'll think about it.

>> The tidiest solution would no doubt be to find an equivalent method in
Zinc, which I am sure exists, and then modify the HTMCSS code to use it. I
have tried to find an equivalent, but Zinc is a large and complex system and
I rapidly got lost. are required by legacy packages, as in this case. Should there at least be a
way of overriding deprecations? (I suppose that is what I have done by
commenting it out, but it seems crude.)

This is how you can work around Deprecation warnings:

[ Url combine: 'http://www.foo.com/one/two/' withRelative: 'bar/readme.txt'
]
  on: Deprecation do: [ :exception | exception resume ]

but that is a temporary hack.

>> Thanks in advance for any help or suggestions.
>>  
>> Peter Kenny
>

Sven

Peter,

> On 10 Jan 2015, at 23:30, PBKResearch <[hidden email]> wrote:
>
>> Yes change the code of the Parser not to use old code.
> As I pointed out, the Parser is a large and complex package, and I am not
> the author, so this is easier said than done. It seems a bit odd to dismiss
> code which worked OK in Moose 4.9, but not in Moose 5.0, as 'old code.'  It
> is code which has no functional equivalent in the new code, as Sven has
> confirmed.

Could you explain what arguments were given to the combine and what the expected output was/should be ? How does it use path merging ?

Thx,

Sven

>> Yes change the code of the Parser not to use old code.
> As I pointed out, the Parser is a large and complex package, and I am not
> the author, so this is easier said than done.
I'm dealing daily with code having the same characteristics.

> It seems a bit odd to dismiss
> code which worked OK in Moose 4.9, but not in Moose 5.0, as 'old code.'

I do not know what happened between 4.9 and 5.0 but if this is between
pharo30 and pharo40 => one full year of efforts (you know
many people improving things and putting energy in the system). And be
ready because we will remove URL old class.
Now the good aspect is that you are lucky (we are all lucky) that sven
is improving constantly Zinc.

But there is no such world were unmaintained packages would
indefinitively work while the world is changing.
>   It
> is code which has no functional equivalent in the new code, as Sven has
> confirmed.
Then it means that you have to cooperate with sven to get Zinc improved.
And feel lucky you can.

The overall process may look frustrating but a system that is not
changing is
See law of software evolution

(1974) "Continuing Change" — an E-type system must be continually
adapted or it becomes progressively less satisfactory[3]

http://en.wikipedia.org/wiki/Lehman's_laws_of_software_evolution

Sven

Apologies for the delay in replying.

Basically, the relative argument can be anything that occurs in a LINK tag
in an HTML document. This can be anything from a file name to a complete
address with only the 'http:' missing.

I have found a couple of examples in a document from the Wikipedia
'Wiktionary', where I show in the attached workspace file the absolute
address, the two relative addresses and the effect of combining using Url
class>>#combine:withRelative and ZnUrl>>#inContextOf:. In each case the
output of the Url class>>#combine:withRelative is the expected result; in
the first case ZnUrl>>#inContextOf: gives the expected result, in the second
it does not.

The relative address in the second case seems a bit odd (why not include the
'http:' and make it absolute?), but it actually occurs (quite often in
Wiktionary files at least) and the code needs to be able to deal with it.

Hope this helps.

Peter Kenny

-----Original Message-----
From: Pharo-users [mailto:[hidden email]] On Behalf Of
Sven Van Caekenberghe
Sent: 10 January 2015 22:56
To: Any question about pharo is welcome
Subject: Re: [Pharo-users] Problem due to deprecation of class Url in Pharo
3 and Moose 5.0

Peter,

> On 10 Jan 2015, at 23:30, PBKResearch <[hidden email]> wrote:
>
>> Yes change the code of the Parser not to use old code.
> As I pointed out, the Parser is a large and complex package, and I am
> not the author, so this is easier said than done. It seems a bit odd
> to dismiss code which worked OK in Moose 4.9, but not in Moose 5.0, as
> 'old code.'  It is code which has no functional equivalent in the new
> code, as Sven has confirmed.

Could you explain what arguments were given to the combine and what the
expected output was/should be ? How does it use path merging ?

Thx,

Sven

Hi Peter,

The problem is that ZnUrl>>#inContextOf: works with 2 parsed URL objects. However, relative URLs are not proper URLs and cannot be parsed nor represented as such. For example, 'foo/bar/readme.txt' and '/foo/bar/readme.txt' are the same.

Relative URLs only exist while parsing and must then be resolved. Section 5 of RFC 3968 explains in details how to do that. This will have to be added to ZnUrl or to a helper class, but that will take some time. Luckily there are test cases in the document.

Thanks for the feedback.

Sven

https://pharo.fogbugz.com/f/cases/14855/Add-reference-resolution-to-ZnUrl