

|
Volume Number: | 6 | |||
Issue Number: | 9 | |||
Column Tag: | Language Translation |
Symbol Tables
By Clifford Story, Mount Prospect, IL
Note: Source code files accompanying article are located on MacTech CD-ROM or source code disks.
A. Introduction
This month, my series on Language Translation returns to lexical analysis, and I present the amazing new, improved Canon tool.
Parts of this tool are identical (or nearly so) to code presented in my third, fourth and fifth installments, and I will not repeat these parts this month (although they are, of course, included on the code disk).
Specifically, the tool is a filter program; I developed a skeleton filter program in my third installment. It uses no fewer than six state machines for lexical analysis and parsing; lexical analysis and state machines were the subject of my fourth part. And it uses the balanced binary tree routines I developed in my fifth part to implement a symbol table.
B. What the Tool Should Do
The Canon tool functions as follows: the program reads in a dictionary of substitutions, then reads input files, performs the substitutions as required, and writes the result. The difference between this Canon tool and the standard MPW Canon is that is will not perform substitutions within comments or strings.
The tool is controlled by the MPW command line. It takes several possible options, which may be in any order.
B(1). The Dictionary File
The dictionary file must be named on the command line, with the “-d <file name>” option. If no dictionary is named, the tool will abort.
The dictionary file’s format is simple: each substitution is specified on a separate line, with the identifier (according to the language’s definition of identifier) to be replaced first, followed by its replacement (which must also be an identifier). For example:
blip blop
specifies that the identifier “blip” should be replaced by the identifier “blop” whereever it occurs.
There is a second form of substitution, which consists of only one identifier. All identifiers in the input that match the dictionary identifier will be replaced by the dictionary identifier. This can be used to force canonical capitalization.
Finally, the dictionary can include line comments. The tool will ignore everything between a ‘#’ sign and the end of the line. It also ignores blank lines.
B(2). The Input Files
Input files may be specified by simply naming them on the command line.
The input files should be either Pascal or C source files. The tool will read them according to their filename extensions: if the file name ends in “.p”, it will be treated as a Pascal file, and as a C file if it ends in “.c”.
If there are several input files, some “.p” and some “.c”, the first one named on the command line controls. If no input file has either a “.p” or a “.c” extension, then Pascal is the default.
If there are no input files named on the command line, the tool will read from standard input. The language will be Pascal.
The “-p” and “-c” options override all of the above language rules and force the language to Pascal or C, respectively. If there are more than one such option specified, the last one controls.
B(3). Other Command Options
The “-o <file name>” option names an output file. If no output file is named, the tool will write to standard output.
The “-s” option will make the tool case-sensitive. The default is case-insensitive.
B(4). Example
Here is an example of the Canon command line:
Canon -d dict file1 file2 -p > dummy
tells Canon to read the input files “file1” and “file2”, performing substitutions from the dictionary file “dict”. The input will be treated as Pascal source, and the output will be written to standard output, which is in turn redirected to the file “dummy”.
C. Designing the Tool
You may have formed the impression that I like table-driven software. This program has no fewer than eight tables in it: two for character translation, one character classification table, four lexical analyzers and a parser. These are all kept in the resource fork.
Driving a program with tables makes the coding simpler. The price you pay is that the logic is hidden in a table, and consequently rather obscure. If you lose your notes, you may have to re-write the whole table to make a minor change! Assuming you hang onto your notes, however, tables make your program easy to change.
After I had written this program, I realized that I had forgotten about strings. Sure, I had a version of Canon that did not make substitutions within comments but it still made them within quoted strings. So I added that at the last minute; I added a few lines and columns to the lexical tables in the resource fork, and changed two constants in the code. That was it.
C(1). Main Routine
The main routine reads the command line, sets appropriate flags, reads the dictionary into the symbol table, and finally filters the input file(s).
It reads the command line in two passes. The first pass is for setting flags; the second does the work. I need to set the flags before reading any files because I need to know the source language before I read the dictionary file.
After the first pass, the routine reads in the dictionary, opens the output file (if any), and then goes into the second pass. The second pass reads and filters each input file, writing the result to the output file (or standard output).
C(2). Case Sensitivity
We want the tool to be case-insensitive unless the command line option -s is used. This will require some modifications to last time’s symbol table routines (the only place where string comparisons occur).
One approach would be to transliterate the key strings before calling “strcmp”. I want to minimize changes to the symbol table routines, though, since I don’t intend to reprint them in this article.
Another way, the way I have chosen, is to write a case-insensitive version of “strcmp”. Then all I need to do is change the name of the call in the “insert” and “lookup” routines.
Probably the most efficient way would be to use the first method in “insert” and the second in “lookup”. Since all the comparisons in “lookup” are between a key string and keys in the table, and the table would already be case-insensitive, I’d need only a “half-case-insensitive” comparison routine for “lookup”.
Of course, I still need to allow for case-sensitive lookup, if the -s flag is set. What I’ll do is have two transliteration tables, one converting uppercase to lower, and the other a straight identity table. I’ll set a global pointer to point to the appropriate table for my comparison routine to use.
C(3). Parsing the Dictionary
The first thing the program has to do is read in the dictionary. It does this in two phases: a lexical analyzer breaks the dictionary into tokens (identifiers, carriage returns, and errors), then a parser finds substitutions in the token stream.
C(3)(a). Lexical Analyser
There are two lexical analyzers, one for Pascal and one for C, because C allows underscores in identifiers. (Another, and probably better, way to do this is to have one lexical analyzer and two character tables.) In the interests of brevity, I will limit the discussion to the Pascal version; the C version is identical, except that it adds “underscore” whereever “letter” appears.
The first piece shows the the pound sign is a line comment character; after reading a pound sign, we scan to the next carriage return and then go back to state 0. We push the return character back onto the input, though, since it isn’t part of the comment.
The second segment reads an identifier. Again, the character that ends the identifier isn’t part of it, so it goes back onto the input. The lozenge thing indicates that we are going to return a token (i.e., accept states). The example lexical analyzer in installment 4 was the whole program, and it never returned anything. This one is called by a parser, and it returns one token each time it is called.
Figure 1: Pascal Dictionary Lexical Analyzer
Finally, the scanner eats white space, returns carriage returns, and if it hits anything else, errors.
Here is the class table:
data ‘TABL’ (1001) { $”00 00 00 05 00 00 00 00 00 04 00 00 00 05 00 00" $”00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00" $”04 00 0D 06 00 00 00 0E 09 0A 0C 00 00 00 00 0B” $”02 02 02 02 02 02 02 02 02 02 00 00 00 00 00 00" $”00 01 01 01 01 01 01 01 01 01 01 01 01 01 01 01" $”01 01 01 01 01 01 01 01 01 01 01 00 00 00 00 00" $”00 01 01 01 01 01 01 01 01 01 01 01 01 01 01 01" $”01 01 01 01 01 01 01 01 01 01 01 07 00 08 00 03" $”00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00" $”00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00" $”00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00" $”00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00" $”00 00 00 00 00 00 00 00 00 00 04 00 00 00 00 00" $”00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00" $”00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00" $”00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00" };
Here is the state table:
data ‘TABL’ (2001) { $”FD 02 FD FD 00 FE 01 FD FD FD FD FD FD FD FD” $”01 01 01 01 01 00 01 01 01 01 01 01 01 01 01" $”FF 02 02 FF FF FF FF FF FF FF FF FF FF FF FF” };
The negative numbers correspond to the lozenges.
C(3)(b). Parser
The dictionary parser is a simple hand-made thing, and does not use YACC (that would be like swatting a fly with a hammer). A dictionary is a list of lines; a line may be blank, or it may contain a one-ID specification, or a two-ID specification. That is,
line -> CR line -> ID CR line -> ID ID CR
Here’s a state machine to implement that grammar:
Figure 2: Dictionary Parser
Recall that the parser gets its data by calling the lexical analyzer, and thus receives only three tokens: ID, CR and ERR. State 3 is the error recovery state; it reads to the end of the line, and then goes back to state 0 for the next line. Returns to 0 from state 0, 1 and 2 correspond to the three lines of the grammar above. In the latter two cases, the specification is added to the symbol table.
Here is the state table:
data ‘TABL’ (1000) { $”01 00 03" $”02 00 03" $”03 00 03" $”03 00 03" };
C(4). Making Substitutions
To make substitutions in the input file, we begin with a lexical analyzer that finds all the identifiers. Again, there are two versions, one for Pascal and one for C. I will discuss the C version only; Pascal does not allow underscores in identifiers, and the two languages have different comment constructs. See the fourth installment of this series for a lexical analyzer that reads Pascal comments.
The first segment reads comments, and is identical to the comment-reader presented in the fourth installment. The next two read strings. (The second segment is also present in the Pascal version, for compatibility, even though Pascal doesn’t use quotation marks for anything.) Canon does not do any syntax checking, and will read strings that go beyond the end of the line.
The fourth segment reads identifiers. When it finds one, the lozenge means “look it up and see if there’s a substitution to be made”. This scanner, unlike the dictionary scanner, doesn’t return anything; it runs until it finds the end of file, making substitutions as appropriate.
Here is the state table (which uses the same class table as the dictionary lexical analyzer):
data ‘TABL’ (3002) { $”00 07 00 07 00 00 00 00 00 00 00 02 00 05 06" $”01 01 01 01 01 00 01 01 01 01 01 01 01 01 01" $”00 00 00 00 00 00 00 00 00 00 00 01 03 00 00" $”03 03 03 03 03 03 03 03 03 03 03 03 04 03 03" $”03 03 03 03 03 03 03 03 03 03 03 00 04 03 03" $”05 05 05 05 05 05 05 05 05 05 05 05 05 00 05" $”06 06 06 06 06 06 06 06 06 06 06 06 06 06 00" $”FF 07 07 07 FF FF FF FF FF FF FF FF FF FF FF” };
Figure 3: C Source Lexical Analyzer
D. Coding the Tool
What follows does not include all the code for the tool. Parts of it are scattered through my last two articles; refer back to those if you need to see it all. Alternately, the entire source is included on the MacTutor source code disk.
// Constants and Macros #define nil 0 #define stdinfd 0 #define stdoutfd 1 #define stderrfd 2 #define stdunit(x) ((x >= stdinfd) && (x <= stderrfd)) #define notstdunit(x)(x > stderrfd) #define nombuffsize1024 #define truebuffsize 1200 #define classcount 15 #define idstate 7 // Types typedef enum {false, true} logical; typedef enum {nocode, pascalcode, ccode} codetype; typedef struct node { char *key; struct node*left; struct node*right; int balance; char *data; } node; // Globals unsigned char *CASETABLE; // Prototypes void initmac(); int openoutput(char *thename, int output); int readinput(int input, Handle inbuffer); int filter(char *inbuffer, int buffersize, int output, codetype language, node *symbols); int writeoutput(int output, char *outbuffer, int buffersize); node *parser(char *dictname, codetype language); int gettoken(char *buffer, int buffersize, char *thestring, char *classtable, char *statetable); node *createnode(char *thekey, char *thedata); unsigned int insert(node *parent, char *thekey, char *thedata, int depth); node *lookup(node *thetable, char *thekey); void destroy(node *thetable); void dump(node *thetable); int compare(unsigned char *string1, unsigned char *string2);
D(1). Main Routine // main int main(int argc, char *argv[]) { int output; logicalsensitive; codetype language; char *outputname; char *dictname; logicalerrors; int index; char *thetail; Handle thehandle; node *symbols; int input; int buffersize; initmac(); // “output” is the fd of the output file, initially stdout // “sensitive” is the case sensitivity, initially insensitive // “language” is the language to parse, initially unknown output = stdoutfd; sensitive = false; language = nocode; // “outputname” is the name of the output file // “dictname” is the name of the dictionary file // “errors” is the error flag, initially indicating no errors outputname = nil; dictname = nil; errors = false; // command line interpreter: first pass for (index = 1; index < argc; index++) { if (argv[index][0] == ‘-’) { switch (argv[index][1]) { // “-p” and “-c” options set // language type; these override // any previous setting case ‘C’: case ‘c’: language = ccode; break; case ‘P’: case ‘p’: language = pascalcode; break; // “-s” option makes Canon case sensitive case ‘S’: case ‘s’: sensitive = true; break; // “-o” option names the output file; // remember the name and read // the file later case ‘O’: case ‘o’: index++; if (outputname == nil) outputname = argv[index]; else errors = true; break; // “-d” option names the dictionary file; // remember the name and read // the file later case ‘D’: case ‘d’: index++; if (dictname == nil) dictname = argv[index]; else errors = true; break; default: errors = true; break; } } else if (language == nocode) { // argv[index] is the name of an // input file; if “language” has // not changed since initialization, // set “language” according to // file name thetail = argv[index] + strlen(argv[index]) - 2; if (compare(thetail, “.p”) == 0) language = pascalcode; else if (compare(thetail, “.c”) == 0) language = ccode; } } // exit if errors were found in the first pass if (errors) exit(2); // if “language” is still unknown, set it to Pascal if (language == nocode) language = pascalcode; // load the case table if (sensitive) thehandle = GetResource(‘TABL’, 4002); else thehandle = GetResource(‘TABL’, 4001); HLock(thehandle); CASETABLE = (unsigned char *) *thehandle; // copy the dictionary into the symbol table if (dictname == nil) exit(2); symbols = parser(dictname, language); if (symbols == nil) exit(2); // if “outputname” has been found, open the output file if (outputname != nil) { output = openoutput( argv[++index], output); if (output < 0) exit(2); } // “input” is the fd of the input file, initially stdin // “thehandle” is the input buffer, initially empty // “buffersize” is the size of “thehandle” input = stdinfd; thehandle = NewHandle(0); buffersize = 0; // command line interpreter: second pass for (index = 1; index < argc; index++) { // skip all options (read in first pass) if (argv[index][0] == ‘-’) { switch (argv[index][1]) { case ‘D’: case ‘O’: case ‘d’: case ‘o’: index++; } } else { // argv[index] is the name of an // input file; open the file and // read it into the input buffer input = open(argv[index], O_RDONLY); if (input < 0) exit(2); buffersize = readinput(input, thehandle); if (buffersize < 0) exit(2); close(input); // call “filter” to read the input buffer // and write filtered output HLock(thehandle); filter(*thehandle, buffersize, output, language, symbols); HUnlock(thehandle); } } // if “input” is still a standard unit // number, then no input file was // opened, and input must be from // standard input if (stdunit(input)) { buffersize = readinput(input, thehandle); if (buffersize < 0) exit(2); // call “filter” to read the input buffer // and write filtered output HLock(thehandle); filter(*thehandle, buffersize, output, language, symbols); HUnlock(thehandle); } // wrapup: dispose of the input buffer, // close “output” if the program // opened it and dispose of the symbol table DisposHandle(thehandle); if (notstdunit(output)) close(output); destroy(symbols); exit(0); }
D(2). Case Sensitivity
This is the string comparison routine to use in place of “strcmp” in the symbol table “insert” and “lookup” routines. The only other change I made to those routines was to rename the local variable “compare” “difference” (to avoid conflicts with this routine name).
The routine functions just like the C routine: it returns a negative number if string1 is less than string2, positive it string1 > string2, and zero if they’re equal. The actual number returned is simply the difference between the first pair of different characters. CASETABLE is a global pointer to the appropriate transliteration table.
// compare int compare(unsigned char *string1, unsigned char *string2) { register int char1; register int char2; register int difference; char1 = *string1++; char2 = *string2++; while (char1 || char2) { difference = CASETABLE[char1] - CASETABLE[char2]; if (difference) return(difference); char1 = *string1++; char2 = *string2++; } return(0); }
D(3). Parsing the Dictionary
I parse the dictionary in two steps: lexical analysis and parsing. The “gettoken” routine breaks the input into tokens, which the “parser” routine fits together into substitution specifications.
D(3)(a). Lexical Analyser
This routine is similar to the lexical analyzer I used in my fourth article. The major difference is that it returns tokens as it finds them, rather than keeping control from the beginning to the end of the file. It knows that it has found a token when it gets a negative state number; it converts it into the token number that the parser expects, and returns it. (This is probably unduly complex; I should have just let the parser use negative token numbers and avoided the conversion.)
// gettoken int gettoken(char *buffer, int buffersize, char *thestring, char *classtable, char *statetable) { static int position = 0; int thestate; unsigned char thechar; int theclass; int newstate; // start the machine in state 0 thestate = 0; while (position < buffersize) { // read the next character, look up its // class, and get the new state thechar = buffer[position++]; theclass = classtable[thechar]; newstate = statetable[classcount * thestate + theclass]; switch (newstate) { // -3 => ERR, -2 => CR; in either case, // just return the the token number case -3: case -2: return(- 1 - newstate); // -1 => ID; return the token number // and the identifier in “thestring” case -1: *thestring = ‘\0’; position--; return(- 1 - newstate); case 0: if (thestate == 1) position--; break; case 1: break; case 2: *thestring++ = thechar; break; } thestate = newstate; } return(-1); }
D(3)(b). Parser
The first half of this routine is set-up work. In addition to loading its own state machine, the parser also fetches gettoken’s state machine. It’s easier to do the work once, here, than to repeat it each time I can gettoken. The it also opens the dictionary file, reads it in, and so on. Eventually, it gets to do some parsing, and this should look familiar.
There is one complication: gettoken will not only return a token number but will, in the case of an identifier, also return the token’s text. I don’t want to overwrite one identifier when I read the next, so I pass a pointer to one string at the beginning of the line, and then a pointer to a second string when I want to read the next identifier.
// parser node *parser(char *dictname, codetype language) { Handle thehandle; char *parsetable; char *classtable; char *statetable; int thefile; int buffersize; char *buffer; node *symbols; int thestate; int newstate; int theline; int errors; int thetoken; char thekey[256]; char thedata[256]; char dummy[256]; char *thestring; // “parsetable” is the parser’s state machine thehandle = GetResource(‘TABL’, 1000); HLock(thehandle); parsetable = *thehandle; // “classtable” is the character class table thehandle = GetResource(‘TABL’, 1001); HLock(thehandle); classtable = *thehandle; // “statetable” is the lexical state machine if (language == pascalcode) thehandle = GetResource(TABL’, 2001); else thehandle = GetResource(‘TABL’, 2002); HLock(thehandle); statetable = *thehandle; // open the dictionary file... thefile = open(dictname, O_RDONLY); if (thefile < 0) return(nil); // and read it into the buffer thehandle = NewHandle(0); buffersize = readinput(thefile, thehandle); if (buffersize < 0) { close(thefile); return(nil); } close(thefile); HLock(thehandle); buffer = (char *)*thehandle; // “symbols” is the symbol table symbols = createnode(“”, “”); // start the machine in state 0, and on line 1 thestate = 0; theline = 1; errors = 0; // read the first identifier into “thekey” thestring = &thekey; thetoken = gettoken(buffer, buffersize, thestring, classtable, statetable); while (thetoken >= 0) { newstate = parsetable[ 3 * thestate + thetoken]; switch (newstate) { // if we got here from state 1, then we // read only one identifier; if from // state 2, we read both “thekey” // and “thedata” // state 0 is the beginning of a line, so // increment the line counter and set // “thestring” to “thekey” case 0: if (thestate == 1) thetoken = insert(symbols, thekey, thekey, 0); else if (thestate == 2) thetoken = insert(symbols, thekey, thedata, 0); if (thetoken == 4) errors++; theline++; thestring = &thekey; break; // having read one identifier into // “thekey”, the next one should go // into “thedata” case 1: thestring = &thedata; break; // having read one identifier into // “thekey”, and the next one // “thedata”, read anything else into // “dummy” case 2: thestring = &dummy; break; // case 3 is the error case; if we just // got here, write an error message case 3: if (thestate != newstate) fprintf(stderr, “”); errors++; break; } thestate = newstate; thetoken = gettoken(buffer, buffersize, thestring, classtable, statetable); } DisposHandle(thehandle); if (errors > 0) { destroy(symbols); return(nil); } return(symbols); }
D(4). Making Substitutions
This routine should be familiar by now, except for when it finds an identifier. The state table flags identifiers with a state of -1; when the routine reaches that state, it looks up the identifier in the symbol table and performs any required substitution. In all other cases (things other than identifiers, or identifiers with no substitution), the routine simply copies the input to the output.
// filter int filter(char *inbuffer, int buffersize, int output, codetype language, node *symbols) { int inposition; int outposition; int thetoken; node *thenode; int thelength; Handle thehandle; char *classtable; char *statetable; char outbuffer[truebuffsize]; int thestate; unsigned char thechar; int theclass; int newstate; int writesize; // “inposition” is the current read position // “outposition” is the current write position // “thetoken” is the position of the // beginning of the current identifier inposition = 0; outposition = 0; thetoken = 0; // “classtable” converts characters into classes thehandle = GetResource(‘TABL’, 1001); HLock(thehandle); classtable = *thehandle; // “statetable” is the state machine if (language == pascalcode) thehandle = GetResource(‘TABL’, 3001); else thehandle = GetResource(‘TABL’, 3002); HLock(thehandle); statetable = *thehandle; // start the machine in state 0 thestate = 0; while (inposition < buffersize) { // read the next character, find its class and the new state thechar = inbuffer[inposition++]; theclass = classtable[thechar]; newstate = statetable[classcount * thestate + theclass]; switch (newstate) { // found an identifier: if it is in the // symbol table, replace it with the // table’s data. Then go to state 0. case -1: inposition--; outbuffer[outposition] = ‘\0’; thenode = lookup(symbols, &outbuffer[thetoken]); if (thenode != nil) { outposition -= strlen(&outbuffer[thetoken]); thelength = strlen(thenode->data); BlockMove((Ptr)thenode->data, &outbuffer[outposition], thelength); outposition += thelength; } newstate = 0; break; // retract if going from state 2 to state // 0; otherwise, copy input to output case 0: if (thestate == 2) inposition--; else outbuffer[outposition++] = thechar; break; // reading an identifier: if this is the // beginning, record the position for // later use. Then, fall through to // the default case idstate: if (thestate != idstate) thetoken = outposition; // all other cases, copy input to output default: outbuffer[outposition++] = thechar; break; } // if the output buffer fills up, and // we’re not in the middle of an // identifier, write it to disk if ((outposition >= nombuffsize) && (thestate != idstate) && (newstate != idstate)) { outposition = writeoutput( output, outbuffer, outposition); if (outposition < 0) return(outposition); } thestate = newstate; } // write the output buffer to disk writesize = write(output, outbuffer, outposition); return(writesize); }
E. Conclusion
The tool, as I have presented it here, is not quite perfect. It is very slow. I ran it using the “cannon.dict” file that comes with MPW; after first finding all the duplicate lines, it took 22 minutes just to load the dictionary! I was stunned.
The problem, it turned out, was the “createnode” routine. There are over 3200 lines in the dictionary file, and “createnode” calls “NewPtr” three times for each line, for a total of almost 10,000 calls to NewPtr. And NewPtr is very slow. When I re-wrote the tool to reduce the 10,000 to a few dozen, the time to load the dictionary dropped to 16 seconds. (Yes, I’m bragging...)
I chose not to present the faster version in this article, because I feel it confuses the issue. The changes I made are not related to the topic, and make the code more complicated. Instead, I’ve included both versions on the source code disk, and I’ll now give a quick description of the differences between the two.
I got rid of two-thirds of the NewPtr calls by leaving the data where I found it. In the above version, I read the file into memory, then find identifiers in the data and copy them into strings, which I pass to “createnode”. Createnode in turn copies these strings into its data structures. In the faster version, I find identifiers in the data and write nulls at their ends, then pass pointers to “createnode”, which simply copies the pointers into the appropriate node fields. So in addition to 6000 NewPtrs, I’ve saved 12,000 string copies.
The complication is writing the null character. There are times when you don’t want to overwrite the following character right away. Suppose it’s a return character...
I reduced the remaining 3000 calls to a handful by allocating the nodes in large arrays. I put new nodes in the free slots of the array until it fills up, with no calls to NewPtr. Once the array is full, I have to use NewPtr to create a new one, but since I use a large array size, this doesn’t happen very often.

- SPREAD THE WORD:
- Slashdot
- Digg
- Del.icio.us
- Newsvine