On Sun, Sep 30, 2012 at 10:15 AM, Sven Van Caekenberghe <sven@stfx.eu>wrote:
Yes, Transcript will be too slow.
Actually, I didn't look carefully at the Deprecation class. It has show/raise warning options and a simple way to collect deprecations with #deprecationsWhile: Maybe I should run production code with
Deprecation raiseWarning: false.
I was going to suggest you that. It is also accesible from the Settings UI. BTW, maybe we should do this in #cleanUpForProduction?
This is of course a global setting.
Your idea with the window sounds cool but how/when are you going to show it ? And what about headless ?
On 30 Sep 2012, at 01:24, Igor Stasenko <siguctua@gmail.com> wrote:
On 29 September 2012 23:12, Sven Van Caekenberghe <sven@stfx.eu> wrote:
Guys,
So yesterday I deployed a new server based on an #20309 image, with a single Metacello config for my code, and a separate manual load for Seaside 3.1 (which I use as an interface to work with the headless image) - and all went well !
But I had to apply one manual hack: remove the deprecation from #includesSubString: (for Seaside).
I feel like we need another, more silent kind of deprecation for situations like this. Library/framework/application writers targetting multiple Pharo versions are being hit way too hard by this, for - in this case certainly -, no good reason. Creating two versions (I introduced a special package Pharo-Forward-Compatibility just for this), is not worth it (unless you absolutely want the best Pharo 2.x code like I do).
I would imagine something similar to Deprecation which maybe prints on the Transcript, but is otherwise silent, except when some switch is turned on.
Putting transcript is possible only for non-performance critical methods. Put it in wrong place, and it will crawl like slug .. So, we need a kind-of non-interfering deprecation.
I am all for having less intrusive deprecation system. IMO, to fix deprecation (if developer really cares), one needs only the method which sends the deprecated message, bringing a full Debugger is overkill in 99% of cases.
What can we do is to make a nice wrapping around this, so if running code hits deprecated method, it should bring a special popup window titled 'deprecated methods used' or something, and if user will click on the button there, he can see the list of methods which having send sites which hit deprecated methods, i.e. a list populated from pairs:
MyClass>>foo using deprecated SystemClass>>deprecatedBar
clicking on that entry will open a browser in #foo method, where you can fix it.
(sure thing, if method #foo runs millions of times, it should not produce millions of very same entries in that list. we only interested in unique sender->receiver pairs. And run-time check for that is far less expensive comparing printing that stuff on transcript every time).
To speed things up even more, we can even just do this on a first hit, and then recompile deprecated method to not do any checks but perform only what they exist for. But this will require some dark magic, either with compiler, or using method wrappers of some sort.
At the end, what we have is a system simply runs as usual, except from small window, which (if users care to click on) will bring the list of deprecations hit so far.
It would give us the power to deprecate much more things even 'silly renames'. And less external stuff would break.
What do you think ?
Sven
-- Sven Van Caekenberghe http://stfx.eu Smalltalk is the Red Pill
-- Best regards, Igor Stasenko.
-- Mariano http://marianopeck.wordpress.com