the-carlisle-group / rumba Goto Github PK

Core.LogHTTPToSession contains this line:

ts←#.CarlisleGroup.Rumba.Core.GetCurrentUTCHttpDate''

Look at X-RateLimit-Remaining, Retry-After, etc. Look at maximum requests per some time period. For example, on each request, have we seen this IP before? If so, increment request count by 1, then check against max, elapsed time, reset, etc.

The help states that NewClient.Host will be set automatically if not set and the value will be derived from Request.URI

I cannot reproduce it but I see in my log files (very rarely) a VALUE ERROR on line 7 of DefaultResponseHeaders

    h,←(~ce)/⊂'Content-Length'(⍕⍵.ContentLength)

This function has a problem:

 PercentDecode←{
     dec←{⎕ML ⎕IO←0
         ⍉16⊥16|'0123456789ABCDEF0123456789abcdef'⍳⍉↑⍵
     }
     v←'%20',⍵
     p←(v≠'%')⊂v
     p2←dec 2↑¨p
     p3←2↓¨p
     b←1↓∊0,¨'+'=p3     ⍝ For backwards compatability we attempt to decode + as ' '
     r←'UTF-8'⎕UCS 1↓⊃,/p2,¨⎕UCS¨p3
     0=+/b:r            ⍝ No blanks: done
     (b/r)←' '
     r
 }

I got this as data:

 content  +%E2%8D%B3%0D%0A    

content should be just ⍳: a space followed by an iota. This causes a LENGTH ERROR.

Generally speaking, when the data carries both a space and a multi-byte character it goes wrong.

... so that the folder may be provided as an argument, and that the left arg to Start can be elided, making it a static method that creates and returns its own server. This allows (along with issue #34 and issue #35):

S←Start 'c:\MyStaticWebsiteFolder'

To be all that is needed to start up a static site that can be accessed by pointing to localhost.

When the IsFile and Path property are used in a Response object to specify the content of the response, a HEAD request returns the content when it should not.

This should point to the Core space.

And one could also argue that Method is wrong

...even though the actual value is in fact UTC.

Function relies on the Windows shell. I will rewrote it to use []MKDIR

and clean up tests.

Not sure if this is cause by an Acre bug, but the folder should go.

 GetCurrentUTCHttpDate←{
...
     _net←⎕NS''
     _net.⎕USING←,⊂'System'
     _net.DateTime.UtcNow.ToString(,⊂'r')

This causes a DOMAIN ERROR.

     _net.DateTime.UtcNow.ToString(⊂,'r')

fixes the problem

from event number to callbacks.

URL such as https://test.expression-builder.com/#/r0/expression-builder/LOANID/if do not drop the trailing / on the result of the function.

 Authority  expression-builder.com                                                      
 Fragment   /r0/expression-builder/LOANID/if                                            
 Host       expression-builder.com                                                      
 Path       /index.html/                                                                
 Port                                                                                   
 Query                                                                                  
 Scheme     https                                                                       
 URI        https://expression-builder.com/index.html/#/r0/expression-builder/LOANID/if 
 UserInfo   m                                                                           

1. ⎕OFF should use a right "argument" that is not 0, indicating an error to the calling process if there is any.

2. It should attempt to log the ⎕OFF under strict error trapping and maybe also ⎕DMX.EM.

3. The function should check whether a Ride session is currently connected and if so not execute ⎕OFF, allowing the user to investigate. If somebody has connected with Ride there is a reason, possibly an interest in finding out what is causing the ⎕OFF.

   Something like this:

 :If 3501⌶0                           
     TRAP←⎕TRAP  ⍝ Remember setting for inspection
     ⎕TRAP←0 'S'
     ∘∘∘                     ⍝ Check it out, dear ⎕AN
 :EndIf

The log file gets created with the name
\Rumba\Log\log.txt
in the parent folder instead of placed in the folder
\Rumba\Log\

I'm in CST at 10:08 am

⎕ts
2018 4 11 10 9 10 747
QTSToHTTPDate ⎕ts
Wed, 11 Apr 2018 10:09:12 GMT

Maybe update the function comments to state that inputs needs to be in GMT?

When requesting the Guardian home page from https://guardian.co.uk/ Rumba crashed because the response from the server has only a length of 2, so the line

     ⍵.ReasonPhrase←↑{⍺,' ',⍵}/2↓m

crashes.

Instead of crashing it seems to be better to check before that line is executed:

  2=≢m:⍵  ⍝ Host response is invalid, and there is nothing we can do about it

.... so the project apce with default to #.Rumba

back to 80 and 0.

The CreateDirectory code for the log file is hardcoded to c:/ windows file system. We should rewrite this to work on an OS independent structure.

.... on both client and server, to accurately reflect what is provided in the message HTTP startline.

RequestTargt rather than RequestTarget

5:VALUE ERROR
GetFormValue[4] c←⍺.ProcessedContent
∧

⍺.ProcessedContent does not exist in the namespace.

The NewTableRowLink codein NewTable should probably look for a zilde and if found, make that row regular non-link text. Right now it makes the link self referencing to the current page. This can be a time consuming round trip with larger data sets.

.... to something that will serve up a static site.

Ability for GetQueryParameter to return a default value if parameter does not exist.

Rumba does not know 422

OpenLogFile creates a log if the Folder property is defined. This poses an issue when pointing to a folder in a read-only or runtime environment. There should me an option to enable or disable, depending on what default is preferred.

... nobody is ever sending a 346?!

... to take a URI and retrieve the page/resource.

When one attempts to start the test server this crashes:

FILE NAME ERROR: C:/.../Rumba/Assets/Runtime/TestCertificates/server/localhost-cert.pem: Unable to open file ("The system can
not find the path specified.")

After copying over the required TestCertificates\ folder from an installation directory TestGetFileRequest crashes.

The Rumba log file is opened pretty late, in particular after the OnStart handler was already executed.

I suggest to open the log as early as possible. I want to log stuff early.

and its inverse ComposeURI, and a self-inverse function URI that covers both.

I'm creating links that have invalid URL characters. I'm torn if I should preprocess them or have the final NewLink method in Rumba properly encode the segments?

I.e.

/v1/directory/APLSource
/v1/directory/TestData
/v1/directory/\Rumba
/v1/directory/\Rumba\Log

The backslashes should be encoded..

Right now it looks like Rumba will replace '' of an HTMLElement with a href to a '/' without encoding it. This breaks the resource.

In case an attempt to establish a connections fails because, say, the server does not respond, then an attempt to clean up the battle field causes a VALUE ERROR in Close[6]:

 Close←{
    ⍝ Public Instance Method
    ⍝ ⍺ ←→ Client
    ⍝ ⍵ ←→ 0
    ⍝ ← ←→ 0
     _←DRC.Close ⍺.Name
     _⎕TKILL ⍺.Tid    ⍝ <==
     ⍺.State←0
     ⍺.Name←''
     ⍺.Messages←⍬
     0
 }

I suggest instead:

 Close←{
    ⍝ Public Instance Method
    ⍝ ⍺ ←→ Client
    ⍝ ⍵ ←→ 0
    ⍝ ← ←→ 0
     _←DRC.Close ⍺.Name
     ⍺.State←0
     ⍺.Name←''
     ⍺.Messages←⍬
     0=⍺.⎕NC'Tid':0   ⍝ When an attempt to open a connection failed then "Tid" might not exist
     _←⎕TKILL ⍺.Tid
     0
 }

Currently the code is swallod by the Connect function.

Although CongaError tries hard to produce something useful as text, in my case that was just "OS Error".

I propose to change CongaError:

 CongaError←{
⍝ ⍵ ← Conga return code.
⍝ ← 1. If rc can be found in DRC.ErrorTable[;0]:
⍝      * If not empty the long error message is returned.
⍝      * Otherwise the error abbreviation is returned.
⍝   2. If ⍵ cannot be found in DRC.ErrorTable  (345 'Unknown Conga error')  is returned.
⍝ The result can be used by:
⍝ ⎕SIGNAL CongaError <errNo>
     rc←⍵
     (abbr long)←(##.DRC.Error rc)[1 2]
     0<≢long:⊂('EN' 345)('ENX'rc)('Message'long)
     ⊂('EN' 345)('ENX'rc)('Message'abbr)
 }

The only difference is that the original return code is available on ⎕ENX.

As a consequence Connect needs to change when calling ConaError:

 Connect←{
...
     rc≠0:⎕SIGNAL CongaError rc
...
 }

To reproduce the bug run this function@

∇ ns←test;stop;ns;R;client;request;response
    stop←{⍺←200 ⋄ ⍺=⍵: ⋄ ⎕←⍵ ⋄ ∘∘∘}
    ns←⎕NS''
    R←#.CarlisleGroup.Rumba
    {}R.Core.InitConga''
    R.Conga.X509Cert.LDRC←R.DRC
    client←R.Core.NewClient 0
    client.Port←443
    client.Secure←1
    client.X509←⎕NEW R.Conga.X509Cert
    client.Host←'api.github.com'
    request←R.Core.NewRequest 0
    request.URI←'https://api.github.com/repos/aplteam/Meddy/releases/latest'
    {}client R.Core.Send request
    response←client R.Core.Receive 0
    {}client R.Core.Close 0
    stop response.StatusCode
∇  ⍝Done

Executing this results in

INDEX ERROR
GatherResponseMessage[13] n←(1+c.Messages⍳⍵)⊃c.Messages

"rootname" allows to use Conga in the same workspace for two different applications, for example one acting as a web server and one acting as a Tatin client. Different "rootnames" make sure that both applications see only their own Conga "objects".

For the Tatin client this is so important that I had to introduce it, so in my Rumba fork InitConga not only accepts a path pointing to a folder that holds compatible Conga DLLs, it also allows specifying a "rootname", both as optional parameters.

I also removed any calls to InitConga from any of the Rumba functions because my changes require Rumba to be initialized before she can actually use it.

With different "rootnames" one has also to use different versions of the Rumba namespace because internally the result is always assigned to DRC. For Tatin that is not posing a problem since it lives in ⎕SE, but there could be a better solution.