Index Page
Error Handling in CSPICE

Table of Contents 


   Error Handling in CSPICE
      Abstract
      CSPICE naming conventions

   Introduction

   Using CSPICE Error Handling
      The First Thing You Should Know
      What CSPICE Does About Errors, Automatically
      Changing What CSPICE Does About Errors
      Choosing Where the Error Messages Are Sent
      Choosing Which Error Messages to Write
         Example 1
         Example 2
         Example 3
         Example 4
         Example 5
      Choosing the Error Response Action
      Handling CSPICE Errors in Your Own Program
      Testing the Error Status
      Retrieving Error Messages
         Getting the short message
         Getting the long message
         Getting the explanation for the short message
         Getting the traceback
      Resetting the Error Status
      Handling Errors Detected in Your Own Program
      Signaling Errors
      Setting the Long Error Message
      Using the RETURN Action and the CSPICE Function, return_c
      Maintaining Traceback Information -- Checking In and Checking Out
      Clean Coding Using failed_c
         Example 1
         Example 2
      Finding Out What the Current Error Handling Settings Are

   Concepts and Definitions
      About Errors
      About Error Messages
         The short error message
         The explanation of the short error message
         The long error message
         The traceback
         The default message
      About Error Response Actions

   Advanced Programming with Error Handling
      Using the CSPICE Functions failed_c and return_c
      Using the CSPICE routines chkin_c and chkout_c
      Using Multiple Settings of the Automatic Error Handling Features


Top

Error Handling in CSPICE




Last revised on 2006 NOV 22 by N.J. Bachman.



Top

Abstract



This Required Reading document details the use of CSPICE error handling facilities.



Top

CSPICE naming conventions



The CSPICE library is an implementation of the FORTRAN SPICELIB library in C. CSPICE is composed of C routines translated to C from FORTRAN by f2c and a set of wrapper functions which allow a more C native interface to the f2c'd routines.

    A routine name which ends in an underscore, "_", is an f2c translated routine (pckopn_).

    A routine name ending in and underscore c, "_c", is a wrapper routine (mxm_c). It is strongly suggested that the user calls a wrapper routine whenever available as opposed to the f2c translated counterpart.

    A routine name in all capital letters (SPKEZR) is a SPICELIB FORTRAN routine.

    Variables in routine calls with the suffix "_len" are the lengths of particular strings in the argument list.



Top

Introduction




Most of this information in this document is OPTIONAL reading, not ``required'' reading. To get started using CSPICE quickly, it suffices to read up through the section titled ``The First Thing You Should Know.''

About the organization of this Required Reading document:

The second chapter describes ``Using CSPICE Error Handling.'' We begin with brief descriptions of the error handling features, and many examples of their use.

The third chapter describes ``Definitions and Concepts.'' This chapter provides a more detailed and complete discussion of the concepts used in describing the CSPICE error handling mechanism, but generally does not give examples. Here we define what we mean by ``errors,'' describe error messages and their use, and describe the possible error processing ``actions'' -- ways of responding to error conditions.

The fourth chapter describes ``Advanced Programming With Error Handling''; that chapter discusses extension of the error reporting scheme to your own application code.

The document covers the following topics:

    -- How to make your program test whether a CSPICE routine has detected an error, and how to find out what kind of error it was.

    -- How to direct CSPICE software to continue running after an error has occurred, rather than stopping.

    -- How to select the messages you want to output when an error is detected.

    -- How to get a log of errors detected by your own application code.

    -- How to direct CSPICE software to send error messages to a file.

    -- How to get a traceback of your call chain when an error occurs in your own code.

Error handling topics beyond the scope of this document:

    -- Information about specific errors detected by CSPICE.

    -- How to take corrective action when an error is detected.

    -- How to debug programs that use CSPICE.

    -- How to determine whether to treat a condition arising in your own code as an error.

    -- How to write code that performs the ``minimum possible'' amount of error checking.

Before starting our discussion of how to use the CSPICE error handling facilities, let's briefly answer the question, ``Why perform error handling at all?'' Here are a few reasons:

    -- It makes your program MUCH easier to debug. Most run-time diagnostics only tell you about low-level types of errors such as arithmetic overflow, attempts to divide by zero, or attempts to write protected memory locations. These run-time errors may occur in very different locations from the original coding mistakes causing them, so the original mistakes are hard to find. CSPICE error handling helps bound the section of code where a coding error is located, greatly simplifying the search for it.

    Also, a good deal of understanding of a program without error handling is often required to fix problems, since problems are diagnosed in low-level terms, not conceptually: it's like the difference between being told that the current through a certain wire in your television is zero, instead of being told that your set isn't plugged in. Understand many mistakes may not cause run-time diagnostics to be produced.

    -- Error handling makes your program much more robust. Your program can respond to errors and continue execution in many cases, with CSPICE error handling.

    -- It makes user-friendliness possible. When errors occur, your program can retain control and present the problem to the user as you choose. Contrast this to an abort with a run-time error message.

    -- It makes error handling portable and uniform across systems. CSPICE error handling is accomplished using standard C. Different systems do not handle run time errors in the same way.



Top

Using CSPICE Error Handling




This chapter tells you how to use most of the features of the CSPICE error handling mechanism. Many examples are given, but details are saved for the next chapter ``Concepts and Definitions.''

The material in this chapter covers three areas:

    -- Controlling CSPICE's automatic error handling features

    -- How to enable your own application to handle errors detected by CSPICE code

    -- How to use CSPICE error handling to respond to errors detected by YOUR OWN code



Top

The First Thing You Should Know



The first thing to know about CSPICE error handling is, you don't HAVE to do anything to use it. As a default, when a CSPICE routine detects an error, CSPICE will print error messages to the default output device (the screen, on most systems), and then STOP.

This capability is built into CSPICE software; you get it automatically when you link your program with CSPICE. No additional action of any kind is needed to make it work.

If this behavior is adequate for your application, you don't need to read the rest of this document.



Top

What CSPICE Does About Errors, Automatically



When a CSPICE routine detects an error (we call this ``signaling'' an error), the CSPICE error handling mechanism will write a set of descriptive error messages to the default output device, which is the terminal screen on most systems. The messages are:

    -- A short, descriptive message.

    -- An expanded form of the short message.

    -- Optionally, a long message, possibly containing data.

    -- A ``traceback,'' showing the sequence of calls leading to the routine that detected the error.

    -- A default message, informing the user that the error handling behavior is adjustable.

After writing these messages, the error handling mechanism will halt execution of the program.

See the section ``About Error Messages'' below for details.



Top

Changing What CSPICE Does About Errors



You can change the set of error messages written when an error occurs. You can re-direct the output to a file, or suppress it. You can choose to have your program continue after an detecting an error rather than stopping.



Top

Choosing Where the Error Messages Are Sent



Suppose you want to have any error messages written to a file, rather than to the default output device (the screen, on most systems). You can do this by calling the CSPICE routine, errdev_c. The first argument should be 'SET'. The second should be the file specification.

For example, to have error messages go to the file, ERROR.DAT, you could use the following code: 

   /*
   Set the error output device for CSPICE to the
   file ERROR.DAT:
   */
   errdev_c ( "SET", lenout, "ERROR.DAT" );
This call should precede calls to other CSPICE routines, except erract_c and errprt_c. If your program has an initialization portion, this call should go there.



Top

Choosing Which Error Messages to Write



By default, when an error is detected by a CSPICE routine, up to five different types of error messages are written. You can tell CSPICE's error handling mechanism to write any combination, including ``all'' and ``none'' of these messages.

Change the set of written messages via a call to the CSPICE routine, errprt_c.

The first argument should be 'SET'.

The second argument is a list of message types that you want to ADD to the set of messages that currently are selected to be output when an error occurs.

The keywords which may be used in the list: 

   SHORT
   LONG
   EXPLAIN
   TRACEBACK
   DEFAULT
   ALL         ( write all messages )
   NONE        ( write no messages  )
The list of message choices is read from left to right, with each word ADDING to the previous set of messages (except for the word NONE, which subtracts all of the messages from your selection).

Some examples may help clarify this.



Top

Example 1



Suppose that currently, the short message has been selected for output. To ADD the long message, make the call: 

   /*
   Add the long message to the set selected for output
   when an error occurs:
   */
   errprt_c (  "SET", lenout, "LONG" );
Now the short and long messages will be output if CSPICE detects an error.



Top

Example 2



What if you want JUST the long message? Put the word NONE at the beginning of the list; this cancels the previous selection: 

   /*
   Just output the long message when an error occurs:
   */
   errprt_c (  "SET", lenout, "NONE, LONG" );


Top

Example 3



What if you want just the traceback and long message? Put the word NONE at the beginning of the list; this cancels the previous selection: 

   /*
   Just output the long message and traceback on error:
   */
   errprt_c (  "SET", lenout, "NONE, TRACEBACK, LONG" );


Top

Example 4



How about no messages? 

      /*
      Don't output ANY CSPICE error messages on error:
      */
      errprt_c (  "SET", lenout, "NONE" );


Top

Example 5



All messages? 

      /*
      Output ALL CSPICE error messages on error:
      */
      errprt_c (  "SET", lenout, "ALL" );
See the section ``About Error Messages'' below for details.



Top

Choosing the Error Response Action



CSPICE's default response to error detection includes halting the program. What if you don't want that?

CSPICE's error handling mechanism allows you to choose one of several different error response ``actions.'' In almost all cases, the only reasonable alternative to the default action is RETURN. Briefly, the full set of choices consists of:

DEFAULT

The action you start with, automatically. This is the same as ``abort'' (below), except that all of the error messages, including the ``default'' message, are written.
ABORT

Print messages and halts the program. With this action, you can select the error messages that are written.
REPORT

Write messages and CONTINUE EXECUTION.
This action may be useful for debugging. Caution: Once an error has occurred, your program may not execute as expected.

RETURN

Write messages, AFTER WHICH ALL CSPICE ROUTINES THAT DETECT ERRORS OR CALL OTHER ROUTINES WILL RETURN IMMEDIATELY UPON ENTRY.
This is the action to use if you want to perform error handling for CSPICE errors in your own code.

IGNORE

DO NOTHING in response to errors. No error messages will be output; the CSPICE status functions return_c and FAILED will not indicate that an error has occurred.
These choices are mutually exclusive; only one action can be in effect at a given time.

You use erract_c to set the error response action. The first argument should be 'SET'. The second argument should be one of the above choices. For example: 

      /*
      Set the SPICELIB error response action
      to "RETURN":
      */
      erract_c (  "SET", lenout, "RETURN"  );


Top

Handling CSPICE Errors in Your Own Program



So far we've talked about how to control automatic reporting of errors detected in CSPICE routines.

The automatic error reporting feature is meant to produce human- readable error diagnostic information. However, you may also wish to have your program respond to CSPICE errors.

To do this, you will need to know about three more basic functions: testing and resetting the CSPICE error status, and retrieving CSPICE error messages.



Top

Testing the Error Status



You use the CSPICE function, failed_c, to tell if any SPICELIB routine has detected an error. failed_c is a function which returns a boolean. It takes the `true' value, if any CSPICE routine has detected an error.

For example, suppose you call the CSPICE routine, rdtext_c, to read a line of text from a file. You want your program to test whether an error occurred on the read. You can write: 

   /* Read a line from userfile.txt; check for errors: */
   rdtext_c ( "userfile.txt", line_len, line, &eof );
 
   if ( failed_c() )
   {
      return;
   }
If you're used to routines that have error arguments, you might note that the code is similar to what you would write if FAILED were an output argument for rdtext_c, instead of a function.

However, there are a number of advantages to the CSPICE method, one of which is that if you don't wish to write any error handling code to handle CSPICE errors, you don't have to, and you'll still get helpful error messages automatically. Also, if you use CSPICE error handling in your own code, you don't need error arguments, which makes for simpler code.



Top

Retrieving Error Messages



CSPICE provides routines to retrieve the error messages generated whenever a CSPICE routine detects an error.

This feature is useful for two reasons. First, if you want your program to take different actions depending on what error occurred, it gives your program a way to find out. Second, if you want to generate your own error reports instead of using those generated by CSPICE, you need to be able to retrieve the information CSPICE has generated about the error.



Top

Getting the short message



Because of its brief format, the short message is the one to use in your code in any logical tests you might want to do.

To retrieve the short error message, call getmsg_c or getsms_. For example: 

   /*
   Call prompt_c to prompt the user for the name of file
   to read from.  Read a line from it. Check for errors:
   */
   prompt_c ( "Enter file name > ", file_len, file );
 
   rdtext_c ( file1, line_len, line, &eof );
 
   while ( !eqstr_c ( file, "" ) )
   {

      if ( failed_c () )
      {
         /*
         An error occurred.
         Find out what the short message was:
         */
 
         getmsg_c ( "SHORT", shrtms_len, shrtms );
 
         if (    eqstr_c( shrtms, "SPICE(NOFREELOGICALUNIT)" )
              || eqstr_c( shrtms, "SPICE(TOOMANYFILESOPEN)"  )  )
         {
            /*
            We won't succeed in reading any file.
            So, quit.
            */
            return;
         }
         else
         {
            /* Get name of a different file: */
            prompt_c ( "Enter file name > ", file_len, file );
         }
      }
   }


Top

Getting the long message



The long error message and traceback aren't useful for program logic, but you may want them if you're going to produce error reports in your own format.

To get the long error message, call getmsg_c or getlms_. For example, 

   rdtext_c ( file, line_len, line, &eof );
 
   if ( failed_c() )
   {
      /*
       Get long message and output it.
      */
      getmsg_c ( "LONG", longms_len, longms);
      printf  ( "%s \n", longms );
   }
The argument supplied to getmsg_c should be declared char *.



Top

Getting the explanation for the short message



The CSPICE routine expln_ can obtain a line of text explaining each CSPICE short error message. This text is an expansion of the short error message, since the short message is frequently abbreviated.

Here's an example: 

   /*
   After this call, EXPL will take the value:
 
   "Invalid Radius--Equatorial or Polar Radius is Zero"
   */
 
   expln_( "SPICE(ZERORADIUS)", expln, 17L, expln_len);
   printf ( "%s \n", expln )


Top

Getting the traceback



Two ways of getting the traceback are provided in CSPICE:

    -- You can call the routine qcktrc_ which returns a character string containing a traceback. Specifically, the string contains a list of module names, separated by arrows. The first name in the string is the name of the highest level module in the active call chain that has ``checked in.''

Example: Suppose MAIN has called SUBA, which in turn has called SUBB. The following code is in SUBB: 

 
   /*
   Get traceback.  After the call to qcktrc_,
   TRACE should have the value,
 
   "MAIN --> SUBA --> SUBB"
   */
 
   qcktrc_(trace, trace_len);
 
   printf ( "%s \n", trace );
    -- You can find out how many routines are in CSPICE's traceback representation by calling TRCDEP (``traceback depth''), and then extract any given name from the traceback representation by calling trcnam_.

For example, you could print out the names of all of the modules in the active call chain with the following code: 

   trcdep_ ( depth );
 
   for ( i=0; i< depth; i++ )
   {
      trcnam_( i , name , name_len);
      printf ( "%s \n", name );
   }


Top

Resetting the Error Status



If your program encounters a recoverable error, you may wish to have it report the error and continue processing.

An example would be the case where you have an interactive program that prompts the user for the name of a file to read. Your program uses the CSPICE function rdtext_c to read the file. If the file isn't found, rdtext_c signals an error. The inability to locate the file need not stop the program; your program could just display a message saying the file wasn't found and ask for another file name.

The problem here is that the CSPICE functions failed_c and return_c will return the `true' value after rdtext_c signals the error, so any code whose logic depends on the value of those functions will behave as if an error has occurred, even though the error was recoverable. To solve this problem, CSPICE provides the routine, reset_c, which ``resets'' the error handling mechanism, so that it acts as if no error had occurred. Calling reset_c has the following effects:

    -- The CSPICE function failed_c will return the `false' value until another error is signaled.

    -- The CSPICE function return_c will return the `false' value until another error is signaled.

    -- getmsg_c, getsms_, and getlms_ will return blank strings.

    -- The traceback routines will show a traceback of the current active call chain, not the active call chain at the time of the last error.

When your program detects a recoverable error, it should fetch the error messages, call reset_c, then complete its response to the error. The diagnostic messages must be fetched before RESET is called. RESET should be called before other SPICE routines are called, because these routines may return upon entry while an error condition exists.



Top

Handling Errors Detected in Your Own Program



When you use CSPICE, you can use the error handling mechanism to respond to errors detected by your own code, as well as by CSPICE code.

Some of the capabilities that you get with CSPICE are:

    -- You can signal errors and have your error messages output to a chosen device, without writing code to do the output.

    -- You can get a traceback of your own call chain, which can be retrieved at any time, not just when an error occurs.

    -- You can control how your program behaves when it encounters an error, just as when CSPICE encounters one.

    -- You can use failed_c as a global status indicator, not just for CSPICE code.

    -- You can use return_c to prevent your routines from executing when an error condition exists, thereby obviating the need to test FAILED after each call to a subroutine.

In the following sections we explain:

    -- How to ``signal'' (indicate) an error condition

    -- How to create error messages

    -- How to use the RETURN error response action and the CSPICE function return_c to help your program handle recoverable errors

    -- How to use the CSPICE routines chkin_c and chkout_c to create a traceback of your own subroutine calls

    -- How to code using failed_c so as to reduce the amount of error checking you need in your code

    -- How to retrieve the settings of the error output device, error response action, and error message selection



Top

Signaling Errors



A routine calls sigerr_c to signal an error condition to the CSPICE error mechanism.

When sigerr_c is called, all of the types of error messages that have been selected for automatic output are written out, and CSPICE takes whatever additional actions are required by the current setting of the error response ``action.''

sigerr_c takes one input argument, a short (25 character maximum) error message. This message will be output if the ``short message'' has been selected for output. It is strongly recommended that your code supply a descriptive (at least non-blank) error message when it calls sigerr_c.

The short message, if used, indicates the type of error which occurred, so the program can respond appropriately.

A capability exists to set a long, human-readable, error message. The next section discusses setting the long message.

Here's an example in which the routine, DACOSH, signals an error. 

      /*
      DACOSH computes an arc hyperbolic cosine of X; X must
      be greater than or equal to 1 to be in the domain of
      DACOSH.
 
      Check that x >= 1.
      */
 
      if ( x < 1. )
         {
         setmsg_c ( "DACOSH: Invalid argument, X is less than one." );
         sigerr_c ( "SPICE(INVALIDARGUMENT)"                        );
         return;
         }
You may note a call to the routine setmsg_c precedes the call to sigerr_c as setmsg_c sets the long error message. setmsg_c is discussed in the next section, but we'll note here that if you wish to call setmsg_c, it should be called BEFORE calling sigerr_c, since sigerr_c causes the current long error message to be output.



Top

Setting the Long Error Message



The long error message is intended to inform the human reader about an error that has occurred.

You may supply a character string of length up to 1840 characters as the input argument to setmsg_c. Strictly speaking, the long message is optional, but it's recommended that you call setmsg_c before every call to sigerr_c, supplying a blank string if you don't wish to set a long message.

As an example, the calls to setmsg_c and sigerr_c from the example in the last section are repeated here: 

    setmsg_c ( "DACOSH: Invalid argument, X is less than one." );
 
    sigerr_c ( "SPICE(INVALIDARGUMENT)"                        );
Frequently, one would like to insert variable strings into a long message. In the above example, it might be nice to convert X, a double precision number, to a character string and put it in the error message. CSPICE provides the routine, errdp_c, for just this purpose.

errdp_c takes two arguments. The first is a character string to appear in the long error message. It marks the place where the result of the conversion is to be placed in the long error message.

The second argument is the value to be converted to a character string. The resulting string is substituted for the first occurrence of the first argument found in the long message.

Here's the previous example, re-written using ERRDP. 

      /*
      Set long error message, with a MARKER where
      the value of X will go.  Our marker is '#'.
      */
 
      setmsg_c ( "DACOSH: Invalid argument, x<1.  The value is #.");
 
      /*
      Convert X to characters, and insert the result
      in the long message where the # is now:
      */
      errdp_c ( "#",  x );
 
 
      /*
      If x happened to be -5.5D0, for example, the long message
      becomes:
 
      "DACOSH: Invalid argument, x < 1.  The value is -5.5D0."
 
 
      Signal the error:
      */
      sigerr_c ( "SPICE(INVALIDARGUMENT)" );
In addition to errdp_c, errint_c and errch_c are provided for inserting integers and character strings into the long message.



Top

Using the RETURN Action and the CSPICE Function, return_c



If you want your program to do any error handling in addition to CSPICE's automatic error handling, you probably would like to prevent your program from crashing as a result of an error, since this unfortunate event may prevent control from ever being given to your error handling code.

CSPICE solves this problem with the boolean function return_c, and the RETURN error response action.

The first two lines of executable code of every CSPICE routine that can detect errors, or that calls another routine, are: 

   if ( return_c () )
      {
      return;
      }
When the error action is RETURN and an error has been signaled, return_c takes the `true' value. So every CSPICE routine that can detect errors, or that calls another routine, returns without doing anything. This greatly reduces the chance of an error causing a program crash.

You can use the function return_c in your own code to achieve the same effect.

return_c always takes the `false' value if the error action is not RETURN.

See the next section to find out what the rest of the IF block should be.



Top

Maintaining Traceback Information -- Checking In and Checking Out



CSPICE can give you a ``picture'' of your call chain at the time an error is signaled, and at any time before. We call this a ``traceback.'' A traceback is provided in the default selection of error messages. When an error is signaled, a traceback helps you find out what calls were made before the call to the routine which detected an error.

As an example, suppose the following figure shows the calling hierarchy for a program, and that currently, subroutine ``E'' is executing, after being called by ``C.'' 

 
        MAIN
      /  |  \
     B   C   D
    / \ /|   |
   E   F E   H
The active call chain would consist of ``MAIN,'' ``C,'' and ``E.'' The traceback message, if retrieved at this point, would be: 

   MAIN --> C --> E
To make your own code participate in the traceback scheme, every routine in your program (except those that call nothing else and don't detect errors) should ``check in'' on entry and ``check out'' on exit. These actions tell the error handling mechanism whether your routine is in the active call chain or not.

To check in, call chkin_c, supplying the name of your routine. To check out, call chkout_c, also supplying the name of your routine. The call to chkin_c should come immediately after each entry into your code. A call to chkout_c should precede each exit made after checking in. For example: 

      /*
      Here's a skeleton of code for a mock routine, suba:
 
      Executable code follows
      */
 
      if ( return_c() )
         {
 
         /* No check out here, since we haven't checked in. */
         return;
 
         }
      else
         {
         chkin_c ( "suba" );
         }
 
               .
               .
               .
 
      if ( x < 1. )
         {
 
         /* First exit following check in: */
         chkout_c ( "suba" );
         return;
         }
               .
               .
               .
 
      /* Normal exit: */
 
      chkout_c ( "suba" );
      return;
The traceback storage can accommodate a stack depth of 100 routines; the maximum length of each stored name is 32 characters.



Top

Clean Coding Using failed_c



When you set the action to RETURN, all CSPICE routines that can detect errors simply RETURN UPON ENTRY. So, they can't cause a run-time error. Therefore, you may safely call a number of CSPICE routines consecutively, and just test failed_c after the last call in the sequence.



Top

Example 1 



   /*
   Read a line from userfile1.txt, userfile2.txt,
   and userfile3.txt; check for errors:
   */
   rdtext_c ( "userfile1.txt", line1_len, line1, &eof );
   rdtext_c ( "userfile2.txt", line1_len, line2, &eof );
   rdtext_c ( "userfile3.txt", line1_len, line3, &eof );
 
   if (  failed()  )
   {
      /* Not all of line1, line2, line3 are valid, so quit: */
 
      chkout_c ( "mysub" );
      return;
   }


Top

Example 2 



   /*
   In the following code, copyc is used to copy the result
   of the union of two sets (ordered cells) from a temporary
   working set back into one of the original sets.
   */
   union_c( &bod,  &planets, &temp );
   copy_c ( &temp, &bod );
 
   if ( failed_c() )
   {
      chkout_c ( "mysub" );
      return;
   }
   /*
   If the size of the temporary cell is greater than the size
   of the original set, failed_c() should be checked to be
   sure that no overflow occurred. If bodies is at least as
   large as temp, no such check is necessary.
   */
You can also use this coding technique with calls to your own routines, if your use the function return_c.



Top

Finding Out What the Current Error Handling Settings Are



Below we explain how to set the error output device, the selection of error messages to output, and the error response action. You can retrieve the current settings of these items by calling the same routines you used to choose the settings, supplying the value ``GET'' as the first argument, instead of 'SET'.

For example, you find out which device error output is sent to: 

   errdev_c ( "GET", lenout, device );
 
   /*
   device, with length lenout, now contains the name of the
   output device.
   */
To find out what the current error response action is: 

   erract_c ( "GET", lenout, action );
 
   /*
   action, with length lenout, now contains the current error
   response action.
   */
To find out what the current error message selection is: 

   errprt_c ( "GET", lenout, list );
 
   /*
   list now contains the last list of messages that
   was input by a call to errprt_c.  If no call was made,
   list has the value "DEFAULT".
   */


Top

Concepts and Definitions






Top

About Errors



What do we mean by ``errors''? In general, we mean ``asking routines to do something they can't do.'' Examples include supplying invalid values of input arguments to a subroutine, exceeding program limits such as the maximum number of files open simultaneously, or trying to read from a file that doesn't exist.

When an error is detected because a routine has been used improperly, information about the context of the error is desirable. It's useful to know in which routine the error was detected, what the call chain was at the time of the error, and what the inputs to the routine that detected the error were. The CSPICE error handling mechanism is designed to provide this type of information.

On the other hand, when it's the program's job to determine the correctness of data, information about the program is not what is wanted when an error occurs. In this case, information about the data is what's needed. CSPICE's automatic error handling is not appropriate for dealing with this type of error. However, it is possible to shut off the automatic error handling, using the IGNORE error action, and use non-CSPICE code to handle these errors.

In general, CSPICE's automatic error handling is most useful for diagnosing programming errors.

The only errors that the CSPICE error handling mechanism deals with are DETECTABLE ones. CSPICE can test whether a calling routine has supplied an argument that's in the domain of a function, but it can't tell if the calling routine has the order of the arguments in a calling sequence reversed. By coincidence, an error may be detected in that case, but the diagnosis will point to the error in an indirect way, at best. And if an application uses a faulty algorithm, but nonetheless uses the CSPICE routines correctly, SPICELIB can't tell you about it.

Some detectable errors exist which CSPICE does not detect. While attempted division-by-zero errors are prevented, floating overflow is generally not prevented (because doing so is too inefficient).

When a CSPICE routine detects an error, it may mean the routine is being used improperly. One of the most likely causes is an interface error: inputs may be supplied to the routine that it can't handle, or there may be an error in the coding of the call to the routine itself. It's a good idea to thoroughly understand the descriptions of inputs and outputs given in the module headers of each CSPICE routine called by one's application.

Some other possible causes of errors may be: bugs in application software, bugs in CSPICE software, or bad inputs to the application program. Errors can also occur due to problems with the program's environment. For example, an attempt to open a file could fail because the application program didn't have the privileges necessary to perform the open, or on some systems, because the file was in use by another user.



Top

About Error Messages



CSPICE uses error messages to inform both the application program using CSPICE, and the user, about what type of error has occurred and where in the program it occurred.

CSPICE provides routines for setting and retrieving error messages. When a routine detects an error, it may ``set,'' or store, error messages, which then can be retrieved and examined by other routines.

There are five types of error messages:

    -- The short error message.

    -- The explanation of the short error message.

    -- The long error message.

    -- Traceback.

    -- The default message.



Top

The short error message



This message is a character string containing a very terse, usually abbreviated, description of the problem. The message is a character string of length not more than 25 characters. It always has the form: 

    SPICE(...)
where the message text goes between the parentheses. An example is: 

    SPICE(FILEOPENFAILED)
The text is always composed of upper case letters and numbers.

Short error messages used in CSPICE are CONSTANT, since they are intended to be used in code. That is, they don't contain any data which varies with the specific instance of the error they indicate.

Because of the brief format of the short error messages, it is practical to use them in a test to determine which type of error has occurred.

For example: 

   rdtext_c ( file, line_len, line, &eof );
 
   if ( failed()_c )
   {
      /*
      An error occurred.  Find out what the short message was:
      */
      getmsg_c ( "SHORT, shrtms_len, shrtms );
 
      if (    eqstr_c(shrtms, "SPICE(NOFREELOGICALUNIT)" )
           || eqstr_c(shrtms, "SPICE(TOOMANYFILESOPEN)"  )  )
      {
          /* We won't succeed in reading any file. So, quit. */
 
          return;
      }
   }
   else
   {
      .
      .
      .
If you use the CSPICE error mechanism to respond to errors detected in your own code, you may wish to use your own short error messages. The CSPICE error handling mechanism doesn't make use of the actual content of the messages, so you may use any values that are practical. It may be of use to make up your own prefix (analogous to CSPICE's ``SPICE'' ), to identify the errors as detected by your own code. Recall the 25-character limit; excess characters will be truncated.

We recommend that you do NOT use blank short error messages. While the error handling mechanism allows it, the short error messages would no longer be useful for enabling code to determine the type of error that has occurred.

The short message is ``set'' by supplying it as an input argument to the CSPICE routine sigerr_c. It is retrieved by calling getmsg_c or getsms_.



Top

The explanation of the short error message



An 80-character expansion of the short message exists for a subset of CSPICE's short messages. Each explanation message is constant. While the short message is meant to be used in code, the explanation is supposed to be easily interpreted by the human reader.

CSPICE provides the routine expln_ to map short error messages to their explanations.

Currently, there is no provision to extend the mapping to user-defined short messages.

In future versions of CSPICE, more space may be allocated for explanations. However, expln_ will continue to return the first 80-character line of the explanation text in that case.

Here's an example of a short message and the corresponding explanation: 

   Short message: "SPICE(ZERORADIUS)"
 
   Explanation:   "Invalid Radius--Equatorial or Polar Radius is Zero"


Top

The long error message



This message may be up to 1840 characters long. The CSPICE error handling mechanism makes no use of its contents. Its purpose is to provide human-readable information about errors.

Long error messages generated by CSPICE routines often contain data relevant to the specific error they describe. Here's an example of a long error message generated by a misspelled body name provided as an input to the CSPICE function spkezr_c: 

   The observer, 'earthh', is not a recognized name for an
   ephemeris object. The cause of this problem may be that
   you need an updated version of the SPICE toolkit. Alternatively
   you may call SPKEZ directly if you know the SPICE ID codes
   for both 'moon' and 'earthh'
The long message is ``set'' by supplying it as an input argument to the CSPICE routine setmsg_c. It may be retrieved by calling getmsg_c or getlms_.

The routines, errdp_c, errint_c, and errch_c provide the capability to insert data into the long message. Their respective purposes are to insert double, int, and char * data into the long message. They (except errch_c) convert the input data to a character string, and insert it in the current long message at a location indicated by a user-specified marker.

We strongly recommend that you DO NOT write code that tests for particular values of CSPICE long error messages; this is a very error-prone practice.



Top

The traceback



The purpose of this message is to represent the active call chain, that is, the set of subroutines in your program that have been called and have not yet returned. The traceback will always show the CSPICE routines in the active call chain.

You can have your own code participate in the CSPICE traceback scheme, so that the traceback will represent the entire active call chain.

Knowledge of the active call chain can be a valuable debugging aid, because it helps you determine where in your program an error occurred. For example, if your program calls subroutine SUBX from ten different subroutines, it's not enough to know that SUBX detected an error; you want to know which subroutine made the offending call to SUBX. The traceback contains that information. Another example: suppose a CSPICE routine in your program detects an error, but your own code does not call that routine. You need to know which call, made by your own code to a CSPICE routine, eventually resulted in the call to the routine that detected the error. The traceback shows you this.

The CSPICE error handling mechanism automatically keeps track of which CSPICE routines are in the active call chain. In order for the CSPICE error handling mechanism to represent the entire active call chain, including non-CSPICE routines, it is necessary for each routine to tell the error handling mechanism: 1) when it has been called and 2) when it is about to return. In CSPICE documentation, these two operations are called ``checking in'' and ``checking out.'' The set of routines that have checked in but have not yet checked out constitute the portion of the active call chain that the CSPICE error handling mechanism can represent.

CSPICE provides the two routines chkin_c and chkout_c for this purpose; chkin_c for checking in, and chkout_c for checking out. chkin_c and chkout_c take one input argument: the name of the calling routine.

The traceback message has the form of a list of subroutine names, delimited by arrows. The first name in the list is the highest-level routine in the portion of the active call chain that is known to the CSPICE error handling mechanism.

As an example, suppose the following figure shows the calling hierarchy for a program, and that currently, subroutine ``E'' is executing, after being called by ``C.'' 

 
        MAIN
      /  |  \
     B   C   D
    / \ /|   |
   E   F E   H
The active call chain would consist of ``MAIN,'' ``C,'' and ``E.'' The traceback message, if retrieved at this point, would be: 

   MAIN --> C --> E
The particular traceback information made available by CSPICE is dependent on the error action and on whether an error has occurred. In general, the routines qcktrc_, trcdep_, and trcnam_ return information about the current active call chain. But when the error response action is RETURN, and an error occurs, the error handling mechanism captures a ``frozen'' copy of the traceback; from this point on, the trace information provided by the above routines will be based on the frozen copy. The purpose of this behavior is to make the traceback that existed at the time the error occurred available to the application program. Changing the error action to a value other than RETURN, or resetting the error status, will cause the traceback routines qcktrc_, trcdep_, and trcnam_ to return data based on the current traceback again.

The overhead of checking in and checking out may be prohibitive for some routines. An intermediate position between not using chkin_c and chkout_c and using them in every routine is to use them only in routines that call other routines, or that detect errors. Routines which do neither would never appear in a traceback anyway. Note that absence of error detection in a routine is not sufficient grounds for exclusion from checking in and checking out, since a routine called by the routine in question could detect an error, and then the caller should appear in the traceback.

It is important to note that ONLY routines which have checked in can possibly have their names appear in the traceback. So the traceback will not be accurate if any routines in the active call chain have not ``checked in,'' when a traceback message is produced.

The traceback mechanism requires any ``checked-in'' routine to check out before exit.



Top

The default message



This message is output when an error occurs, and the error response action is DEFAULT. It informs the user that the behavior of CSPICE error handling is user-tailorable, and it refers the user to this document.



Top

About Error Response Actions



The term ``error response action'' refers to the action that the CSPICE error handling mechanism takes automatically when an error is signaled. CSPICE provides the routine erract_c to enable you to select the error response action.

There are five different error response actions, each appropriate to a different type of programming situation:

DEFAULT

This is what you get if you don't choose an error response action. When the error action is DEFAULT, and an error is signaled, the CSPICE error handling mechanism outputs all of the available types of error messages to the error output device, and then stops program execution.
The default action is probably acceptable as a response to detection of program bugs. However, if you wish to allow your own program to respond to errors, you must choose one of REPORT, RETURN, or IGNORE as the error response action.

ABORT

This action is the same as the default, except that the set of error messages is selectable. This action is useful for running programs in batch mode, to prevent them from continuing execution after encountering an error. The ABORT action is also useful if your program does not perform any error handling itself.
REPORT

This action can be useful for debugging. With the REPORT action, the CSPICE error handling mechanism outputs error messages when an error is signaled, but doesn't stop your program. So, more information can be gained from a single run of your program than with the ABORT action. Using the CSPICE routine errdev_c, you can direct the error messages output by the CSPICE error handling mechanism to a file, thus creating a log of your errors.
However, the function failed_c will indicate an error condition after an error has been signaled. Therefore, any program logic that depends on the value returned by failed_c may be affected by an error. You can use the CSPICE routine reset_c to set the error status to ``no error,'' so that your program will attempt to follow a normal thread of execution, after an error is signaled.

RETURN

The RETURN action's intend use is in situations where you want your program to recover from errors detected by CSPICE (or by your own code). Like the REPORT action, the RETURN action also includes writing error messages to the chosen output device. But with the RETURN action, the CSPICE function return_c takes the `true' value when an error is signaled. Any routine that tests the function return_c on entry, and that returns immediately if return_c takes the `true' value, will not crash in this case. All CSPICE routines that can detect errors, or that call other routines, do test return_c on entry and return if the `true' value. This behavior maximizes the chance that CSPICE routines will return control to the application program using them, in the event of an error.
It is also possible to use the function return_c in the same way in your own routines. We note here that use of return_c mode allows you to call multiple CSPICE routines consecutively, without testing the error status until after the last call. This can greatly enhance the readability of your code, and increase its efficiency, as well.

Returning control to the non CSPICE portion of the program allows for a response to the error condition. The nature of the response is application-dependent. However, in the case that the error does not preclude further successful operation of the program, it is necessary to instruct the CSPICE error handling mechanism that normal operation is desired, after the error has been responded to. The CSPICE function RESET is provided for this purpose. Calling RESET blanks out all stored error messages, and causes the functions FAILED and return_c to take the `false' value.

IGNORE

This action is provided to allow you to completely ``shut off'' CSPICE's automatic error handling. When this action is selected, and any routine signals an error, no action is taken by the CSPICE error handling mechanism. The error messages cannot be updated by sigerr_c and setmsg_c, and return_c and FAILED will not change their values.
This action may be useful in the case where you know exactly which errors will be detected by CSPICE routines, and you know that no response is required.

Note that the IGNORE action is not appropriate for the case where you wish to suppress automatic output of CSPICE error messages, and to handle CSPICE errors in your own code. To do that, use the RETURN error action and set the selection of error messages to 'NONE', using errprt_c.



Top

Advanced Programming with Error Handling




This chapter discusses some further error handling topics.



Top

Using the CSPICE Functions failed_c and return_c



First, note that the CSPICE function return_c takes the `false' value unless the error response action is RETURN, so all of the following discussion assumes that the error action has been set to RETURN.

Which routines should use the function return_c, and where?

The ``which?'' question is somewhat tricky: there's a trade-off between speed and resistance to program crashes. It takes time (not much) to test return_c, and in a very short routine, such as one that computes a dot product, the execution time can be increased by a substantial factor if return_c is tested. In most routines, the percentage increase in execution time is small.

So, if you don't want to test return_c in every routine, how do you decide? The CSPICE answer to this question is, test return_c in routines that call other routines or detect errors (CSPICE routines that are specifically intended to perform error handling are exempt from testing return_c). If a routine calls another routine already, it's unlikely that testing return_c will slow its execution intolerably. Similarly for routines that test for error conditions.

Our final answer is still a hedge: the proper use of return_c depends on your speed and reliability requirements. The CSPICE method may be a good zero-order approximation to what's optimal.

NOTE: You must be very careful about using return_c in application code that DOES error handling, otherwise your error handling code itself may not function when an error is signaled.

The ``where?'' question also eludes a straightforward answer. The idea behind the RETURN action is to ensure that control is returned to whatever part of your program handles errors, when an error is signaled. According to this idea, once an error has been signaled, you have to do something about the error before you can finish the job. So, routines that do use return_c ought to test it at the very beginning of their executable code; there's no reason for them to proceed further until the error condition has been acted upon.

What about using the function return_c in other parts of the code, for instance following calls to CSPICE routines? The same speed vs crash resistance trade-off applies, but there is another consideration: return_c has an effect only when the error action is RETURN. So, in cases where an error condition makes it likely that a portion of code won't work, it's usually better to test the function failed_c, rather than return_c, since a program crash is usually undesirable, even when using the REPORT action for debugging. But if you do want to retain the option of executing the code when an error has already occurred, use return_c.

We repeat here that the CSPICE function failed_c indicates whether an error has been signaled, regardless of the error response action. It can be used to avoid executing any code that may cause a run-time error if an error has already occurred. Note that testing failed_c instead of return_c means that whether the error response action is RETURN or REPORT doesn't affect the program logic in question; the logic depends on whether or not an error has been signaled.



Top

Using the CSPICE routines chkin_c and chkout_c



The simplest approach to the use of these routines is to use them in every routine you write. But as with the case of return_c, your routines will be slightly slower if they call chkin_c and chkout_c, and for very short routines, the percentage difference may be great.

The CSPICE approach is to call chkin_c and chkout_c in routines which call other routines or which can detect errors. This way, every routine that can possibly be in the active call chain at the time an error is detected can appear in the traceback message.

As mentioned in the previous section, these routines are less likely to suffer a large percentage increase in execution time as a result of making extra calls.

We note that some routines that do make external references but do not signal errors may have no routines capable of signaling errors below them in the calling hierarchy; these routines won't appear in a traceback. Why should they call chkin_c and chkout_c? They should do so for maintenance reasons, specifically, if any routine below them in the calling hierarchy is changed so as to be able to detect an error, the higher level routines don't need to be changed.



Top

Using Multiple Settings of the Automatic Error Handling Features



In the second chapter, we suggested that you choose the error output device, error response action, and selection of error messages, during program initialization and leave them unchanged for the duration of the program run. This is probably a good procedure to stick to for inexperienced users. However, it's possible and reasonable to do some fancier things with these settings, in some cases:

    -- Resetting the output device. This isn't particularly fancy; you may simply want to write error messages from different parts of your program to different devices. errdev_c may be called from anywhere in your program, as many times as your wish. Note that the error response action should not be DEFAULT or ABORT, for in this case the first error signaled terminates execution.

    -- Resetting the error message selection. Nothing to it, just call errprt_c any time. Again, note that the error response action should not be DEFAULT or ABORT, for in this case the first error signaled terminates execution.

    -- Resetting the error response action. erract_c can also be called at any time. However, it is important to be aware of the special effects of these error response action ``transitions.'' Specifically:

    When making a transition to a new action, the value of failed_c does not change. If failed_c has the `true' value, and you set the action to IGNORE, any NEW errors that are signaled will be ignored, but failed_c will remain `true'.

    The value of the function return_c depends on FAILED and the current action. Specifically, return_c has the `true' value if and only if failed_c has the `true' value and the action is RETURN. Therefore, note that if failed_c has the `true' value and you change the action to RETURN, the function RETURN will now have the `true' value, regardless of its previous value. Setting the action to something other than RETURN will cause return_c to take the `false' value, regardless of its previous value.

    Changing the error action doesn't affect the stored short and long error messages, if any. If there are stored messages, they will still be available if the action is changed to IGNORE.

    The traceback routines return information about the ``frozen'' traceback if and only if return_c has the `true' value. So the rule about which traceback you get is the same as the rule for the value of return_c.

    Changing the action to ABORT or DEFAULT doesn't automatically abort your program if FAILED has the `true' value; an error must be signaled while the action is one of these values to achieve an abort.

    -- Temporary resetting of the error handling selections. What if you want to set the error action to ABORT or IGNORE while a certain section of code executes, and then set it back to its original value? Easy: call erract_c with the ``GET'' option to get the current error action. Save this value, call erract_c with the ``SET'' option to set the new value, and finally set the action back to the saved value.

For example: 

   /*
   All errors in the following section of code are
   fatal.  Temporarily set the error action to
   "ABORT"; restore old action after end of code
   section.
   */
 
   erract_c ( "GET", len_savact, savact  );
 
   erract_c ( "SET", len_savact, "ABORT" );
 
 
    { Section of code in which errors are fatal starts here }
 
                          .
                          .
                          .
 
 
    { Section of code in which errors are fatal ends here }
 
 
   /*
   Restore old error action:
   */
   erract_c ( "SET", len_savact, savact );
Changes to the error output device or error message selection can be made in the same way; both errdev_c and errprt_c accept the values ``GET'' and ``SET'' for their first argument.