Re^3: WWW::Mechanize - error on geting non-existing page

[Limbic~Region]

Your Mother,
I honestly can't think of any reason this might be true. Do you also establish a DBI connection without RaiseError set to true? It is akin to not checking the return code of open. If something goes wrong, I want to know about it very loudly instead of blindly assuming everything is ok.

Can you give me an example of where setting it to false is better? I assure you I haven't adopted this strategy as a cargo cult practice and am genuinely interested in hearing your views on how this is better.

[Your Mother]

…well, all examples really. Though it depends on how you see it. I never advocated ignoring return states.

Not finding a page online is not an exception(al case). It’s entirely normal. Should your car blow-up if a road-sign is missing or should the driver revisit the map? If you need to deal with any of the various 400s and 500s, it’s just easier on the hacker to not wrap it up in evals. The end code will have identical functionality: i.e., every request is checked for success or unexpected codes and then processed accordingly. All the exception handling buys you here is additional complexity (and the chance to fall into eval problems with disappearing $@ if you’re not careful).

The change to make this stuff fatal by default is recent-ish. For most of the life of Mech, it wasn’t this way so it was a pretty big style break for those of us using it from the beginning, and I think most LWP hackers are already accustomed to checking $response->is_success habitually.

There may be some value in the fatals for the newcomer who has incorrect expectations but for the experienced user of the kit it’s just an annoying hoop (and a bunch of broken scripts when the update came through but I digress… uh, and whine).

[GrandFather]

I hear where you are coming from, but it sounds a little like "We didn't have warnings in the past, we always checked using defined. Why should we switch to using them now?". To me the difference is between having to check everywhere and maybe getting bitten bad if you missed something, versus getting a noisy failure if something unanticipated happens.

In most cases I'd rather have a noisy failure and deal with it than a quiet failure that may just plough on and destroy the world.

[Your Mother]

Let me put it differently. Mech is a browser emulator. How sensible would it be for a browser to quit on 404s? Mech is a subclass of LWP::UserAgent. How sensible is it to change default behavior radically? Both are obvious mistakes.

This particular change was supposed to help new users, one imagines, but I’ve seen this question come up constantly since it was made: What the heck is going on? My script just quits!? And for adept users it broke all existing scripts and forces a new line of code in everything. I love the functionality of being able to put in your own code ref in but having it be the default is a mistake: conceptually, historically, and expectation-wise.

[Limbic~Region]

Your Mother,
If instead of:

You can also just set onerror => undef which is easier and less error prone than eval/try style code.

You had said:

This behavior can be turned off by either setting onerror to undef or by turning autocheck off in the constructor. The purpose behind this was the author's attempt to protect new users from themselves. Just as you shouldn't assume that your call to open was successful, you shouldn't assume your call to get was either. It will be your responsibility to check each success for is_success/status and handle any recoverable situations yourself if you do this.

I would never have commented. Personally, I like having the code blow up to remind me of places I haven't properly checked the status and leave it on by default but I can see your point as well.