#LyX 1.1 created this file. For more info see http://www.lyx.org/
\lyxformat 2.16
\textclass literate-article
\language default
\inputencoding default
\fontscheme default
\graphics default
\paperfontsize default
\spacing single 
\papersize Default
\paperpackage a4
\use_geometry 0
\use_amsmath 0
\paperorientation portrait
\secnumdepth 3
\tocdepth 3
\paragraph_separation indent
\defskip medskip
\quotes_language english
\quotes_times 2
\papercolumns 1
\papersides 1
\paperpagestyle default

\layout Title

LyX and Literate Programming
\newline 
An example program
\layout Author

Edmar Wienskoski Jr.
\newline 
edmar-w-jr@technologist.com
\layout Date


\latex latex 

\backslash 
today
\layout Standard


\begin_inset LatexCommand \tableofcontents{}

\end_inset 


\layout Section

Introduction
\layout Standard

After typesetting a document, LyX scans the LaTeX log file looking for errors.
 For each error found, the line number is obtained and a error box is displayed
 in the LyX screen at that position.
\layout Standard

To use this feature to view compilation errors while working with literate
 documents, we need a program that filters the compilation errors and puts
 them in a format suitable for LyX reading it.
 
\layout Standard

In this document we present a filter that recognizes compilation error messages
 from noweb, gnu C, and the IBM C compiler (xlc).
\layout Standard

The filter is required to read from standard input, parse for error messages
 and copy the error messages to the standard output.
 During the output process, the filter must present the error messages in
 a format that LyX can interpret, currently, the LaTeX error message format.
 Of course, nothing will prevent future LyX releases from being able to
 read other formats as well (like gcc error messages for example).
 This mechanism is necessary to fully explore the literate programming tool's
 capabilities.
\layout Section

Algorithm
\layout Scrap

<<Function bodies>>=
\newline 
int
\newline 
main (int argc, char **argv)
\newline 
{
\newline 
  if (argc == 2) {
\newline 
    switch (argv[1][0]) {
\newline 
    case 'n':
\newline 
      <<Scan input for noweb error messages>>
\newline 
      break;
\newline 
    case 'x':
\newline 
      <<Scan input for xlc error messages>>
\newline 
      break;
\newline 
    case 'a':
\newline 
      <<AIX system using both noweb and xlc>>
\newline 
      break;
\newline 
    case 's':
\newline 
    case 'b':
\newline 
      <<Solaris and Linux systems using both noweb and gcc>>
\newline 
      break;
\newline 
    case 'g':
\newline 
    default:
\newline 
      <<Scan input for gcc error messages>>
\newline 
      break;
\newline 
    }
\newline 
  } else {
\newline 
    <<Scan input for gcc error messages>>
\newline 
  }
\newline 
}
\newline 
@
\layout Scrap

<<Function prototypes>>=
\newline 
int main (int argc, char **argv);
\newline 
@
\layout Section

Data Structures
\layout Standard

We resort to some global variables to allow access from several different
 routines.
 These are the buffer and related pointers used during the parse of the
 input.
\layout Scrap

<<Global variables>>=
\newline 
char    buffer[200][200];
\newline 
int     last_buf_line;
\newline 
int     last_err_line;
\newline 
int     err_line;
\newline 
@ 
\layout Section

The output format
\layout Standard

The output format mimics the LaTeX error messages format.
 This function prints a number of lines residing in the global variable
 
\family typewriter 
buffer
\family default 
, a program name and line number.
 There is no special requirement on the input strings, they can be anything.
\layout Scrap

<<Function bodies>>=
\newline 
void
\newline 
output_error (int buf_size, int error_line, char *tool)
\newline 
{
\newline 
  int     i;
\newline 
 
\newline 
  fprintf(stdout, "! Build Error: ==> %s ==>
\backslash 
n", tool);
\newline 
  fprintf(stdout, "l.%d ...
\backslash 
n", error_line);
\newline 
  for (i=0; i<buf_size; i++)
\newline 
    fprintf(stdout, "%s", buffer[i]);
\newline 
  fprintf(stdout, " ...
\backslash 
n");
\newline 
}
\newline 
@
\layout Scrap

<<Function prototypes>>=
\newline 
void output_error (int buf_size, int error_line, char *tool);
\newline 
@
\layout Section

Functions Implementation
\layout Standard

Both noweave and notangle routines, always output one single line for each
 error found, thus to scan the buffer for noweb error messages is enough
 to exam one input line at a time.
 Note that the noweb software does not provide a line error number, so all
 errors boxes related to noweb messages will be displayed at the beginning
 of the file.
\layout Scrap

<<Scan input for noweb error messages>>=
\newline 
{
\newline 
  last_buf_line = 0;
\newline 
  while (fgets(buffer[0], 200, stdin)) {
\newline 
    if (noweb_try(0))
\newline 
      output_error(1, 0, "noweb");
\newline 
  }
\newline 
}
\newline 
@
\layout Standard

The examination itself is very inefficient.
 Unfortunately noweb doesn't have any characteristic that would help to
 identify one of its error messages.
 The solution is to collect all possible output messages in an array of
 strings, and turn the examination process into a linear search in this
 array.
\layout Scrap

<<Global variables>>=
\newline 
char *noweb_msgs[] = {
\newline 
  "couldn't open file",
\newline 
  "couldn't open temporary file",
\newline 
  "error writing temporary file",
\newline 
  "ill-formed option",
\newline 
  "unknown option",
\newline 
  "Bad format sequence",
\newline 
  "Can't open output file",
\newline 
  "Can't open temporary file",
\newline 
  "Capacity exceeded:",
\newline 
  "Ignoring unknown option -",
\newline 
  "This can't happen:",
\newline 
  "non-numeric line number in" 
\newline 
};
\newline 
@
\layout Standard

A noweb error message can be any string that contains a matching pair of
 < <\SpecialChar ~
\SpecialChar ~
\SpecialChar ~
> >, or any of the above strings.
\layout Scrap

<<Function bodies>>=
\newline 
int
\newline 
noweb_try (int buf_line)
\newline 
{
\newline 
  char    *s, *b;
\newline 
  int     i;
\newline 
 
\newline 
  b = buffer[buf_line];
\newline 
  s = strstr(b, "<<");
\newline 
  if (s != NULL) {
\newline 
    s = strstr(s+2, ">>");
\newline 
    if (s != NULL)
\newline 
      return 1;
\newline 
  } else {
\newline 
    for (i=0; i<12; i++) {
\newline 
      s = strstr (b, noweb_msgs[i]);
\newline 
      if (s != NULL)
\newline 
        break;
\newline 
    }
\newline 
    if (s != NULL)
\newline 
      return 1;
\newline 
  }
\newline 
  return 0;
\newline 
}
\newline 
@
\layout Scrap

<<Function prototypes>>=
\newline 
int noweb_try (int buf_line);
\newline 
@
\layout Standard

The xlc compiler always outputs one single line for each error found, thus
 to scan the buffer for xlc error messages it is enough to exam one input
 line at a time.
\layout Scrap

<<Scan input for xlc error messages>>= 
\newline 
{
\newline 
  last_buf_line = 0;
\newline 
  while (fgets(buffer[last_buf_line], 200, stdin)) {
\newline 
    if (xlc_try(0))
\newline 
      output_error(1, err_line, "xlc");
\newline 
  }
\newline 
}
\newline 
@
\layout Standard

A xlc error message is easy to identify.
 Every error message starts with a quoted string with no spaces, a comma,
 a space, the word 
\begin_inset Quotes eld
\end_inset 

line
\begin_inset Quotes erd
\end_inset 

, a space, and some variable text.
 The following routine tests if a given buffer line matches this criteria:
\layout Scrap

<<Function bodies>>=
\newline 
int 
\newline 
xlc_try (int buf_line)
\newline 
{
\newline 
  char    *s, *t;
\newline 
 
\newline 
  t = buffer[buf_line];
\newline 
  s = t+1;
\newline 
  while (*s != '"' && *s != ' ' && *s != '
\backslash 
0')
\newline 
    s++;
\newline 
  if (*t != '"' || *s != '"' || strncmp(s+1, ", line ", 7) != 0)
\newline 
    return 0;
\newline 
  s += 8;
\newline 
  err_line = atoi(s);
\newline 
  return 1;
\newline 
}
\newline 
@
\layout Scrap

<<Function prototypes>>=
\newline 
int xlc_try (int buf_line);
\newline 
@
\layout Standard

The gcc compiler error messages are more complicated to scan.
 Each error can span more than one line in the buffer.
 The good news is that every buffer line on each error has the same pattern,
 and share the same line number.
 Thus the strategy will be to accumulate lines in the buffer while the reported
 line number is still the same.
 At the time they differ, all the accumulated lines, except the last one,
 will belong to one single error message, which now can be output-ed to
 LyX.
\layout Standard

Every gcc error message contains a string with no space followed by a 
\begin_inset Quotes eld
\end_inset 

:
\begin_inset Quotes eld
\end_inset 

.
 If the next character is a space, then this line is a header of a error
 message and the next line will detail the line number of the source code
 where the error was found.
 Otherwise, the next thing is a integer number followed by another 
\begin_inset Quotes eld
\end_inset 

:
\begin_inset Quotes eld
\end_inset 

.
\layout Scrap

<<Scan input for gcc error messages>>=
\newline 
{
\newline 
  char    *s, *t;
\newline 
 
\newline 
  last_buf_line = 0;
\newline 
  while (fgets(buffer[last_buf_line], 200, stdin)) {
\newline 
    /****** Skip lines until I find an error */
\newline 
    s = strpbrk(buffer[last_buf_line], " :");
\newline 
    if (s == NULL || *s == ' ')
\newline 
      continue; /* No gcc error found here */
\newline 
    do {
\newline 
      <<gcc error message criteria is to find a "...:999:" or a "...: ">>
\newline 
      /****** OK It is an error message, get line number */
\newline 
      err_line = atoi(s+1);
\newline 
      if (last_err_line == 0 || last_err_line == err_line) {
\newline 
        last_err_line = err_line;
\newline 
        continue; /* It's either a header or a continuation, don't output
 yet */
\newline 
      }
\newline 
      /****** Completed the scan of one error message, output it to LyX
 */
\newline 
      discharge_buffer(1);
\newline 
      break;
\newline 
    } while (fgets(buffer[last_buf_line], 200, stdin));
\newline 
  }
\newline 
  /****** EOF completes the scan of whatever was being scanned */
\newline 
  discharge_buffer(0);
\newline 
}
\newline 
@
\layout Scrap

<<gcc error message criteria is to find a "...:999:" or a "...: ">>=
\newline 
/****** Search first ":" in the error number */
\newline 
s = strpbrk(buffer[last_buf_line], " :");
\newline 
last_buf_line++;
\newline 
if (s == NULL || *s == ' ') 
\newline 
  <<No gcc error found here, but it might terminate the scanning of a previous
 one>>
\newline 
/****** Search second ":" in the error number */
\newline 
t = strpbrk(s+1, " :");
\newline 
if (t == NULL || *t == ' ')
\newline 
  <<No gcc error found here, but it might terminate the scanning of a previous
 one>>
\newline 
/****** Verify if is all digits between ":" */
\newline 
if (t != s+1+strspn(s+1, "0123456789")) 
\newline 
  <<No gcc error found here, but it might terminate the scanning of a previous
 one>>
\newline 
@
\layout Scrap

<<No gcc error found here, but it might terminate the scanning of a previous
 one>>=
\newline 
{
\newline 
  err_line = 0;
\newline 
  discharge_buffer(1);
\newline 
  continue;
\newline 
}
\newline 
@
\layout Standard

As we mentioned, when the scan of one gcc error message is completed everything
 in the buffer except the last line is one single error message.
 But if the scan terminates with a EOF or through finding one line that
 does not match the gcc error message criteria, then there is no 
\begin_inset Quotes eld
\end_inset 

last line
\begin_inset Quotes erd
\end_inset 

 in the buffer to be concerned with.
 In those cases we empty the buffer completely.
\layout Scrap

<<Function bodies>>=
\newline 
void
\newline 
discharge_buffer (int save_last)
\newline 
{
\newline 
  if (last_err_line != 0) {
\newline 
    if (save_last != 0) {
\newline 
      output_error(last_buf_line-1, last_err_line, "gcc");
\newline 
      strcpy (buffer[0], buffer[last_buf_line-1]);
\newline 
      last_err_line = err_line;
\newline 
      last_buf_line = 1;
\newline 
    } else {
\newline 
      output_error (last_buf_line, last_err_line, "gcc");
\newline 
      last_err_line = 0;
\newline 
      last_buf_line = 0;
\newline 
    }
\newline 
  }
\newline 
}
\newline 
@
\layout Scrap

<<Function prototypes>>=
\newline 
void discharge_buffer (int save_last);
\newline 
@
\layout Standard

To combine the scan of noweb error messages and xlc error messages is very
 simple.
 We just try each one for every input line:
\layout Scrap

<<AIX system using both noweb and xlc>>=
\newline 
{
\newline 
  last_buf_line = 0;
\newline 
  while (fgets(buffer[0], 200, stdin)) {
\newline 
    if (noweb_try(0))
\newline 
      output_error(1, 0, "noweb");
\newline 
    else if (xlc_try(0))
\newline 
      output_error(1, err_line, "xlc");
\newline 
  }
\newline 
}
\newline 
@
\layout Standard

To combine the scan of noweb error messages and gcc error messages is simple
 if we realize that it is not possible to find a noweb error message in
 the middle of a gcc error message.
 So we just repeat the gcc procedure and test for noweb error messages in
 the beginning of the scan:
\layout Scrap

<<Solaris and Linux systems using both noweb and gcc>>=
\newline 
{
\newline 
  char    *s, *t;
\newline 
 
\newline 
  last_buf_line = 0;
\newline 
  while (fgets(buffer[last_buf_line], 200, stdin)) {
\newline 
    /****** Skip lines until I find an error */
\newline 
    if (last_buf_line == 0 && noweb_try(0)) {
\newline 
      output_error(1, 0, "noweb");
\newline 
      continue;
\newline 
    }
\newline 
    s = strpbrk(buffer[last_buf_line], " :");
\newline 
    if (s == NULL || *s == ' ')
\newline 
      continue; /* No gcc error found here */
\newline 
    do {
\newline 
      <<gcc error message criteria is to find a "...:999:" or a "...: ">>
\newline 
      /****** OK It is an error, get line number */
\newline 
      err_line = atoi(s+1);
\newline 
      if (last_err_line == 0 || last_err_line == err_line) {
\newline 
        last_err_line = err_line;
\newline 
        continue; /* It's either a header or a continuation, don't output
 yet */
\newline 
      }
\newline 
      /****** Completed the scan of one error message, output it to LyX
 */
\newline 
      discharge_buffer(1);
\newline 
      break;
\newline 
    } while (fgets(buffer[last_buf_line], 200, stdin));
\newline 
  }
\newline 
  /****** EOF completes the scan of whatever was being scanned */
\newline 
  discharge_buffer(0);
\newline 
}
\newline 
@
\layout Section

Wrapping the code into a file
\layout Scrap

<<listerrors.c>>=
\newline 
#include <stdio.h>
\newline 
#include <strings.h>       
\newline 
 
\newline 
<<Global variables>>
\newline 
<<Function prototypes>>
\newline 
<<Function bodies>>
\newline 
@
\layout Standard

To build this program, we want to add the 
\begin_inset Quotes eld
\end_inset 

-L
\begin_inset Quotes erd
\end_inset 

 option in the tangle command to force gdb to load the file 
\family typewriter 
Literate.nw
\family default 
 instead of 
\family typewriter 
listerrors.c
\family default 
.
 In accordance with this, we pass the 
\begin_inset Quotes eld
\end_inset 

-g
\begin_inset Quotes erd
\end_inset 

 option to gcc.
\layout Scrap

<<build-script>>=
\newline 
#!/usr/local/bin/bash
\newline 
notangle -L -Rlisterrors.c Literate.nw > listerrors.c
\newline 
gcc -g -o listerrors listerrors.c
\newline 
@
\layout Standard

This project can be tangled and compiled from LyX if you set 
\family typewriter 

\backslash 
build_command
\family default 
 to call a generic script that always extracts a scrap named 
\family typewriter 
build-script
\family default 
 and executes it.
 Here is a example of such generic script:
\layout LyX-Code

#!/bin/sh
\newline 
notangle -Rbuild-script $1 | sh
\the_end