FIRST

The following pseudo code is relevant in the context of Bugs 3923, 3924, 3989, 4058, 4005.
I tried to elaborate more for the conditions that seemed more probable to be faulty.

PDFAssessment: parse() {

// SeparateQuestions()

// AssociateImages()

// CombineSplitQuestions()

// RecognizeQuestions()

}

Pseudo code for SeparateQuestions():

for each string block "i":

1) Find new section or part

2) In the text block, if found 

(i)  delimited question

(ii) "answer key" text

(iii)instr block and range

(iv) image tag

(v)  digit 

- if expected question # found => set nextNumMatches and 
          update qnum list and the index (multi-digit case)!
        - else if new section found
        - else if answer key found {} (goto for "i")!
        - else append text to current question {} (goto for "i")!

- if correct number found then

(below are examples of cases handled where
             134 is an arbitrary question number)

exclude 134" " and 134"a" 
            exclude 134."a" and 134.")" and 134."$" 
            exclude 134"  " and 134" a"

allow: 
            134)     
            134" ".    
            134" "$
            134#
            134 a
            134 #

- check for answer key "<number>) <letter>" 
            - if index+1 is digit then append to current Q
            - else set current Q's question number and prepend
                  instr block if any!

- else if new page then change pagenum and set question to current 
        - else if left or right indented text
            - if right then append

(vi) instr block 

- update instr block with current
        - Remove the text from list and i--.

(vii)white lines between two consecutive lines of text

- if text corresponds to instructions (e.g., questions i-j)
          then choose the text to be the new question text and set
          indented to false.
        - else append current text to prev question and add new box
          to existing one. Remove the text and i--.

(viii)else same question but check indentation

- if left indented and choice text found (e.g., a. or b) )
          then add the current text to prev question and update box.
        - else set indented to true if current.x > q.x.                              
          Add the current text to prev question and update box. 
        - Remove text and i--.

Most of the cases are handled by this last condition! (e.g., choices, multiple
lines of question). Infact, this is what results in bug 4058 due to faulty indentation check!

Original Bugzilla ID was 4058