ERROR DETECTION AND RECOVERY

8/31/2012

CONTENTS

8/31/2012

ERROR
Program submitted to a compiler often have errors of various kinds So, good compiler should be able to detect as many errors as possible in various ways and also recover from them (i.e) even in the presence of errors ,the compiler should scan the program and try to compile all of it.(error recovery).

 when the scanner or parser finds an error and cannot proceed , the compiler must modify the input so that the correct portions of the program can be pieced together and successfully proessed in the syntax analysis phase. Therefore , there should be a considerable forethought from the beginning in designing the compiler.

 various ways: After the detection of a error, * a simple compiler may stop all the activities other than lexical and syntactic analysis. * a more complex compiler may transform the erroneous input into a similar legal input on which the normal processing can be resumed(repair).

 *an even more sophisticated compiler may correct the erroneous input by guessing what the user has intended. However, no compiler can do true correction.Because,compiler wont know the intent of the programmer due to errors. Completely accurate error correction can be done only by the programmer.

DIAGNOSTIC COMPILER
Compilers giving due importance to all the 3 aspects is called Diagnostic compilers, which follows complex analysis task, which may require extra time or memory space. Eg: WATFOR, IITFORT

CORRECTING COMPILER
These compilers does the job of error recovery not only from the compiler point of view but also from the programmers point of view(ie) generates code to be executed, which eases the programmer. Eg:PL/C But at the same time, error recovery should not lead to misleading or spurious error messages elsewhere (error propagation).

 Indication of run time errors is another neglected area in compiler design. Because, code generated to monitor these violations increases the target program size, which leads to slow execution. So these checks are included as debugging options(also includes intermediate display of values, trace of procedure calls) at the cost of additional amount.

Good error reporting plays an important role in the construction of reliable programs. PROPERTIES OF GOOD ERROR DIAGNOSIS: *The messages should pinpoint the errors in terms of original source program rather than in some internal representation, which is unknown to the user. *The error messages should be tasteful and understandable by the user. Eg: Showing error as missing right parenthesis in line 5 rather than as a cryptic code OH17 *The messages should be specific and should localize the problem Eg: Showing error as Z not declared in procedure add rather than missing declaration. *The massages should not be redundant. Eg: If z is not declared, then it should be said once not every time z appears in the program.

PROPERTIES OF GOOD ERROR DIAGNOSIS

*The messages should pinpoint the errors in terms of original source program rather than in some internal representation, which is unknown to the user. *The error messages should be tasteful and understandable by the user.

 Eg: Showing error as missing right parenthesis in line 5 rather than as a cryptic code OH17 *The messages should be specific and should localize the problem Eg: Showing error as Z not declared in procedure add rather than missing declaration. *The massages should not be redundant. Eg: If z is not declared, then it should be said once not every time z appears in the program.

SOURCES OF ERROR
* ALGORITHMIC ERRORS: The algorithm used to meet the design may be inadequate or incorrect *CODING ERRORS: The programmer may introduce errors in implementing the algorithms, either by introducing logical errors or using the programming language constructs improperly

 *The program may exceed a compiler or machine limit not implied by the definition of the programming language. Eg: An array may be declared with too many dimensions to fit in the symbol table , an array may be declared with too large to be allocated at runtime. *COMPILER ERRORS: compiler can insert errors as it translates source program into an object program.

 CRETERIA FOR THE CLASSIFICATION OF ERRORS: *Compile time, *Link / Load time * Run time errors.

 CLASSIFICATION OF COMPILE TIME ERRORS: *lexical errors * Syntactic errors * Semantic errors.

 The lexical and syntactic errors are found during the execution of the program. Most of the run time errors are semantic in nature. In compile-link-go systems, the compile and link errors will be trapped seperately by the compiler and the linkage editor / loader. In compile-and-go systems, the compile and link errors will be trapped by the compiler itself.

 In compile-and-go systems, the compile and link errors will be trapped by the compiler itself. Execution time errors are detected by the run time environment, which includes runtime control routine, the machine hardware and the standard OS interfaces through which status of the hardware can be accessed or monitored as when required.

PLAN OF ERROR DETECTION IN PORTION OF COMPILER

Its consists of routine to recover from lexical and syntactic errors , a routine to detect semantic errors and a routine to print the diagnostics The diagnostic routine communicates with the symbol table to avoid printing redundant messages. The message printer must defer its diagnostics until after complete line of source text has been read and listed.

8/31/2012

19

ERRORS SEEN BY EACH PHASE

Each phase of the compiler expects its input to flow in certain specification When the input does not, the phase has detected an inconsistency or error which it should report to the user. Moreover in order to continue processing its input, phase has to recover from each error as being lexical phase , syntactic phase or semantic phase errors depending on which compiler phase detects them.
8/31/2012 20

 After detecting and reporting error , a phase can either repair it or pass it along to the subsequent modules. If a phase attempts repair , it should take precautions that the repair does not introduce a flurry of other errors If a phase transmits error the subsequent phase should be able to deal with the erroneous inputs passed on.

8/31/2012

21

PLAN OF ERROR DETECTOR / CORRECTOR

Diagnostic Message Printer
Symbol Table Lexical corrector Syntactic corrector

Source code

Lexical analyzer

Tokens

Parser

Intermediate code

Semantic checker
22

8/31/2012

LEXICAL AND SYNTAX ERRORS

Two frequent sources of these errors are: 1.Spelling errors, 2.Missing operators and keywords These errors can happen due to genuine oversight or due to typing mistake.

They are common mistakes for even for professional programmers.

SPELLING ERRORS- WHEN DO THEY OCCUR???

If a program uses variables names which differs in only one or two characters ,then there exits great scope for spelling errors. There is less chance for using automatic procedures for detecting and correcting these errors. Only the programmer can able to tackle the problem because he only know about significance of variable names.

MAJORITY SPELLING ERRORS

1. 2. 3. 4. One character is wrong, One character is missing, One character is extra, Two adjacent characters are transposed.

Testing for these four types of error will not enable us to catch all the spelling mistakes but practical consideration limit searches to these four only
The implementation these four checks is quite expensive because an associative search has to be performed over all names in symbol table to locate resembling name

CORRECTION ALGORITHM
The searching try to mask off one or more adjacent characters from symbols and locate a matching symbol from symbol table. The located symbol can be used instead of erroneous symbol if only one character was masked off If two characters were masked then transposition should be checked . If associative search for an erroneous names matches more than one symbol in the symbol table then In that case their attributes are used to decide the final. If more than one symbol with matching attributes result then correction is not safe and should not be attempted. This algorithm may fail if unusual usage of names results from valid usage of language facilities. So it is necessary to inform the user whenever one correction made.

MISSING OPERATORS AND KEYWORDS

It can detected by their context. It is not perfect because certain context tends to hide the absence of an operator . ex: G=H(A+B) typed instead of G=H*(A+B).

In this case only if H exist in the symbol it shows error .

Otherwise the couldnt produce any error since H could later be declared to be a function or an array.

DUPLICATE MASSAGE
It is to find that many message appear owing to the same error. Ex: If a is used as a simple variable and later goes to declare and use it as array a[10,10] then all references to the array will be flagged as erroneous use of variable name. This can achieved by setting a flag in the symbol table entry of a. This will enable to detect and indicate all possible illegal use of identifier.

RECOVERING FROM SYNTAX ERROR

The chief concern while recovering from the syntax error is to attain a parser state from where the parser can safely resume parsing the input string. Many parsers detects errors when it doesnot have legal move from its correct configuration , which is determined by its state , stack content and current input symbol. To recover from an error a parser should ideally locate it , correct and resume parsing

TIME OF DETECTION VALID PREFIX PROPERTY

LL1 AND LR1 parsers will announce errors as soon as the prefix of the input has been seen for which there is no valid continuation This is the earliest time at which a parser that reads its input from left to right can announce an error Adv reports errors as soon as possible Limits amount of erraneous output

8/31/2012

30

Panic mode recovery

Parser discards input symbol until a synchronizing token usually a statement delimiter or semicolon is found The parser then deletes stack entries until it finds an entry that can continue parsing given the synchrnosing token on input Ie.skip until we encounter a symbol which tells us what should be the parser state inorder to recognize it Adv simple to implemement Never go infinite loop

8/31/2012

31

on discovering an error the parser discards input symbols one at a time until is found one of a designated set of synchronizing tokens
delimiters ; or } have a clear and unambiguous role must be selected by the compiler designer

skips considerable amount of input no checking for additional errors simple guaranteed not to go on an infinite loop

8/31/2012

32

Three basic policy of recovering syntax error: 1.Deletion of a source symbol 2.Insertion of a synthetic symbol. 3.Replacement.

8/31/2012

33

 The motive behind all these actions is to present a new string to the parser which would lead to bypassing the error situation and continue to parse. Here multiple recovery possibilities may exists. We should choose the one which has smallest number of changes minimum distance recovery
8/31/2012 34

RECOVERY IN TOP DOWN PARSING

There are two methods. First is to try and successfully complete the predictions existing in the stack at the error point. Ex: Input string: A Last prediction was W:AB If no other rules exit with A on right hand side then recovery can be effected by inserting B and deleting parts a until a is recognized in source string. Another is unstack certain symbols from parser stack until we have a TOS symbol which can produce one of the synchronizing symbols. We can skip until we find a synchronising symbol in input stack

RECOVERY IN BOTTOM UP PARSING

In bottom up parsing insertion of symbols is better than deletion. Because it is easy to determine what symbol is to be inserted . routines may be devised to carry out of the specific recovery action. Replacing or deleting the next few source symbols also done.

OPERATOR PRECEDENCE PARSING

Operator precedence parser uses set of production rules and operator precedence table to parse an arithmetic expression.
EE+E |EE |E*E |E/E |E^E |(E) |-E | id

ERROR RECOVERY IN OPERATOR PRECEDENCE PARSING

There are two types of operator precedence parsing errors. character pair errors reducibility errors. A character pair error occurs when there is no operator precedence relation between pairs of symbols in the grammar. A reducibility error occurs when you cannot reduce the handle to the left hand side of some production.

CHARACTER PAIR ERROR RECOVERY

Fill each empty entry with a pointer to an error routine.
Example,

E1 missing operand whole expression is missing E2- unbalanced right parenthesis E3- missing operator E4- missing right parenthesis

REDUCIBILITY ERROR RECOVERY

Decides the popped handle looks like which right hand side. And tries to recover from that situation.
Same like shift-reduce errors

HANDLING SHIFT-REDUCE ERRORS

Generic shift-reduce strategy: If there is a handle on top of the stack, reduce Otherwise, shift

But what if there is a choice? If it is legal to shift or reduce, there is a shift-reduce conflict If it is legal to reduce by two different productions, there is a reduce-reduce conflict

HANDLING SHIFT-REDUCE ERRORS

Ambiguous grammars always cause conflicts But beware, so do many non-ambiguous grammars To resolve this, we should modify the grammar.

SEMANTIC ERRORS
Can be both local and global in scope. Types Immediate errors Can be detected while processing the erroneous statement itself. Delayed errors Cant be detected while processing the statement. But can be detected at a later stage when its effect is felt.

EXAMPLES FOR SEMANTIC ERRORS

Illegal Operator or Operand Control Structure Violation Missing Labels Duplicate Labels (immediate) (both) (delayed at the end of the program)

THE ERROR PRINT ROUTINE

Messages have to be displayed for all errors which are detected , or detected and corrected in the source program. The error print routine is the common agency that is used by all individual compiler routines for this purpose. The text of the error is normally stored in the table local to this routine Associated with each message is the numerical value indicating its error severity This value is mainly used for purposes internal to the compilers operation ( like if not to allow the program to reach the execution stage or not)

1 2 3 4

Warning and correction. Compilation continues and the compiled program will execute Warning only. Compilation continues and compiled program will execute Fatal error. Compilation continues but the compiled program will not execute Compiler error . Compilation terminated

 For each individual error two items of information need to be passed to this routine The error number and the statement number The structure and logic of the routine depends largely on the decisions regarding the place where the message is to be printed.

Desirable place for printing error messages

The messages are best printed against the erraneous statement itself Single pass compilers find it difficult to indicate all errors against the offending statement Multipass compilers can provide such error condition

 Many Fortran compilers indicate errors on a line by line basis as far as possible since syntax analysis and output listing are both performed in the same normally first pass Some compilers group all error messages at end of the program. This has the advantages that the problem of duplicate messages for similar misuse of an identifier can be satisfactorily solved.

 The compiler error table will be in the form

Error number(Message identifier) Erroneous statement Auxiliary information

Message text

Runtime errors
The runtime errors are detected by 1. The run time control routine which is interfaced with the generated code in standard manner 2. The machine hardware 3 . Operating system interfaces to I/O

 The agency required to detect particular type of error depends on nature of error and in general varies from machine to machine and compiler to compiler

Detection of runtime errors

Arithmetic exceptions Arises because of the violations of semantics of machine computations. Includes frequently occurring error conditions like overflows, underflows, loss of precision etc.. Present day architecture detects most of the conditions at the machine hardware level and indicate their presence through interrupts or traps

8/31/2012

53

 Input output errors Device error conditions and end of file conditions on input file are usually detected by IOCS which sets appropriate flag to indicate their occurance The runtime control routine should make appropraite provisions to obtain control when such conditions arise Ex: fortran programmer stmt read(5,100,err=110,end=120) A,B Appropriate code take control to line 120 on eof and 110 on error
8/31/2012 54

 Dimensions overruns overall array bound check Individual subscript bound check Watfor compilers do these

8/31/2012

55

Programmer Recovery Options

Difference b/w compile time and run time error is the type of recovery possible and its implications for pgmr. Syntax errrors can be patched up in a standard manner in order to extend the life of the program and to push it to exeution. Same thing is for runtime errors but here the difference is that the programmer can forsee the runtime errors and correct it Standard recovery action may not suit for a programmer. languages provide this options. PL/I. When ever an exception occurs the runtime control routine has to decide what action to take. Maintains runtime exception table

 Ex. ON SUBSCRIPTRANGE I = 5; ON OVERFLOW I = 25;

I = A(J,K-4)/X; Type of Exception Overflow Program Action I = 25 System Action Make the standard assumption regarding the resulting value. Return. Cancel the Program.

Subscript Range

I=5

 Compiler generates code for inserting and deleting entries from the program action fields depending on the scope of the program-indicated recovery actions. Type of Exception Scope Recovery action stack pointer Overflow -------------------Subscript range ------------------- Scope column indicate where the scope of the programmer indicated recovery action ends.

Debugging aids and options

Run time checks are so costly in terms of code space and execution time. This checks are debugging options.

Trace and Sub traces.

Procedure calls printed out at user option to indicate the flow of control. The trace is written into special debugging files. Debugging file consist of output for the statements and variables.

Assignment Checks
Assignments to a variable are monitored by the system. If the new value is assigned to a variable, which will be the output.

Intermediate and error Dumps

Intermediate dumps can be produced during the execution time. It may also be produced at abnormal program execution.

Conversational debugging
Facilities are provided through which the programmer can set the break points in the program. When the program reach the break point, a conversation is initiated with the programmer. Ex, Assigining values and Displaying the values.

QUERIES??

Slide:61