Exception in callback of async function: SyntaxError: Unexpected token ^~H

[andycaramba]

Hi there. I have this error in logs instead of phantomjs error message.

Exception in callback of async function: SyntaxError: Unexpected token ^~H
    at Object.parse (native) 
    at packages/jazeee:spiderable-longer-timeout/lib/server.coffee:169:22
    at packages/jazeee:spiderable-longer-timeout/lib/server.coffee:33:2
    at runWithEnvironment (packages/meteor/dynamics_nodejs.js:108:1)

My phantomjs version is

~$ phantomjs --version
1.9.0

[jazeee]

Out of curiosity, what platform are you on?

[andycaramba]

Ubuntu 14.04.3 LTS 64 bit and Meteor 1.1.0.3

[jazeee]

Looks like an issue with json parsing. PhantomJS seems to be a bit odd with what it outputs, and appears to put errors on stdout.

You can try version 1.2.2 of this package which does not parse JSON, and should work, albeit with fewer functions.

I use a very similar platform, and don't see these errors. I think we will need to add some debugging logs on parse errors so that we can find the issue.

[andycaramba]

Ok, i'll try to log raw phantomjs message and write results here.

[jazeee]

I pushed an update v1.2.7 which should log it.

Edit: Tested for normal cases which works. It should work to catch the exception. (I don't have exceptions, currently)

[andycaramba]

Oh, thanks. I will try to deploy my app with this new version and observe errors. It would be better to get rid of exception throwing in this block, IMHO. Just logging to console quite enough I think.

[jazeee]

Ok, Rearranged a tiny bit to ensure it goes down the failure path.

Edit: v1.2.8

[andycaramba]

I have caught the error.

Failed to parse PhantomJS output from:  {"status":301,"headers":[{"name":"Server","value":"nginx"},{"name":"Date","value":"Tue, 18 Aug 2015 22:42:38 GMT"},{"name":"Content-Type","value":"text/html"},{"name":"Content-Length","value":"178"},{"name":"Connection","value":"keep-alive"},{"name":"Location","value":"URL HERE"}],"content":"RIGHT BUILDED HTML PAGE HERE"}

I think html content may contain some invalid characters for json parsing. But that's not phantomjs error. Html content is right builded page. Only status 301 confused me. With original meteor package it returns status 200.

[jazeee]

Odd. I took that string and tried it on Chrome console. It is fully valid from what I can tell. I also tested the regular expression, which works as expected. On Chrome console, JSON.parse(...) works as expected, returning an object. I don't understand how it failed to parse or do the regex.

Also, the 301 seems quite odd. Which version of Spiderable did you use. We used to have an issue where the return code was 301, since it was based on all file requests, not just the main route. I wouldn't expect 301 with v1.2.6 or newer.

[andycaramba]

I was looked carefully at html content contained in this error and found this problem symbol which breaks json parsing - <U+2028>.

And I don't remember exactly Spiderable version but it was a version supplied with meteor 1.1.0.2 and 1.1.0.3 at least.

[jazeee]

Where is that character in reference to the JSON? Is it part of the content block? If so, can you paste a trimmed version of that content block to help debugging? Ideally, your test block should fail when running something like:

JSON.parse({"content": "text<U+2028>test"})

[jazeee]

Note, that character appears to be a line separator: http://www.fileformat.info/info/unicode/char/2028/index.htm

[andycaramba]

Yes, it's part of content block. And when i trying to run JSON.parse({"content": "text<U+2028>test"}) in chrome js console i get an error

Uncaught SyntaxError: Unexpected token o(…)

[jazeee]

Sorry, I missed a quote. It should be:

JSON.parse('{"content": "text<U+2028>test"}')

[jazeee]

I am looking to find out how to insert your bad character in a string.

[andycaramba]

I think this SO answer is explain the error - http://stackoverflow.com/a/9168133/585470

[andycaramba]

JSON.parse('{"content": "text<U+2028>test"}') yes it works. I also overlooked the missing quotes.

[andycaramba]

JSON.parse('{"content": "text\u-2028test"}') this code throws an exception

[jazeee]

\u format is \u2028. You can test by: test='{"content":"\u2010"}' The odd thing is this still works:

JSON.parse('{"content":"a\u2028b"}').content.length

[andycaramba]

Oh, sorry. My mistake. Yes, it works with \u2028

[jazeee]

I believe we can remove the problem characters using something like:

test='\u2028';
test.length; // is 1.
test.replace(/\u2028/, '').length; // Is 0

[jazeee]

Probably best on the phantomJS side, to replace all HTML content with \u2028 and \u2029 with spaces? Would that work for your content?

[andycaramba]

I think it would. But the question remains why phantomjs considers it an error. And confusing 301.

[jazeee]

I believe the error is occurring on the Meteor side (in node). Phantom executes successfully, and returns JSON, as shown in your logs. What fails is that JSON.parse. I am not sure it is caused by \u2028, though, since JSON.parse seems to work. For example, our tests still show that this works: JSON.parse('{"content":"a\u2028b"}').content.length

Perhaps there is a different character in your content that is failing. The original error says "^H", which may be a backspace, like http://www.fileformat.info/info/unicode/char/0008/index.htm

[jazeee]

Ahh. It is not an error on Node, or on Chrome console. It is an error when running on PhantomJS:

$ phantomjs 
phantomjs>  JSON.parse('{"content":"a\u2028b"}')
Parse error
phantomjs>  JSON.parse('{"content":"a"}')
{
   "content": "a"
}

Whereas, on Node:

$ node
>  JSON.parse('{"content":"a\u2028b"}')
{ content: 'a
b' }
> 

[andycaramba]

The first error text I've copypasted here was from multitail tool and it seems this tool such interprets this symbol. I've just found problem string in multitail output and it is SOME LONG-LONG TEXT ^~H\\n\\n

Next output I've copypasted was from less tool and it shows this symbol more clear.

[jazeee]

I also tested the same using the node version that Meteor uses, and had no issues. Only PhantomJS executable seems to have the problem. Here is the issue, however. I cannot create the regex in PhantomJS. This fails:

phantomjs> test.replace(/\u2028/g,' ')
Parse error

[jazeee]

Somehow, PhantomJS cannot handle the special character at all. In your test, do you know what \u code that &~H character is? How can we verify that character is 2028?

[andycaramba]

I'm not exactly sure in this but I see this unicode combination with "less" tool in my server remote terminal.

[andycaramba]

I will try to fix it on my side, just making sure that those characters are no longer present on the pages and see what happens.

[andycaramba]

I apologize for such a long answer. I've got rid from this character in html pages and the problem was gone. Thanks for your help and such useful plugin.