rgr@m10ux.UUCP (Duke Robillard) (04/06/88)
In article <764@ndsuvax.UUCP> ncreed@ndsuvax.UUCP (Walter Reed) writes: >Here is an example of why these things bother me. > IF (x = TRUE) THEN > WriteString("Error #"); > WriteCard(e); > WriteString(" in line #"); > WriteCard(l); > WriteString(" detected."); > END; I really like modula2. It's probably my favorite language. But this WriteCard, WriteString stuff makes me want to pop someone in the nose. It's as if someone told Prof. Wirth "heh, pascal's io is pretty bad, but I bet you could make it EVEN WORSE." I hate C as much as the next guy, but it seems to me that printf is the solution. Has anyone ever written some kind of front end for a modula compiler to turn printf's into WriteSh*t's? It doesn't seem like it would be too hard, although you probably couldn't do it with a macro processor like m4...could you? -- + | Duke Robillard | AT&T Bell Labs m10ux!rgr@ihnp4.UUCP | Murray Hill, NJ {any biggy}!ihnp4!m10ux!rgr
bds@lzaz.ATT.COM (BRUCE SZABLAK) (04/08/88)
In article <535@m10ux.UUCP>, rgr@m10ux.UUCP (Duke Robillard) writes: > I hate C as much as the next guy, but it seems to me that printf is > the solution. I like C++'s overloading of the << and >> operator's even better than printf. It allows you to define custom print routines for each structure (class) declared, and then to print (to stdout for example) you do: cout << god_awful_structure_instance; The advantage of this approach over using procedures in modula2 or pascal is (besides previty) that you don't have to remember the type of structure your variable is; the compiler figures it out.
avi@taux01.UUCP (Avi Bloch) (04/11/88)
> >I hate C as much as the next guy, but it seems to me that printf is >the solution. Has anyone ever written some kind of front end for a >modula compiler to turn printf's into WriteSh*t's? Another, more general solution, is the one we took when developing National's Modula-2 compiler. We wanted to be able to interface with any C routine, even those that weren't as strict as Modua-2 routines (e.g. num. & type of parameters procedure or function, etc.). The following is an excerpt from the manual on the extension we added for this purpose: PSEUDO-MODULE 'C' ----------------- A pseudo-module 'C' has been implemented which, like the pseudo-module SYSTEM does not actually exist but is built into the compiler. Module C is used to import C routines that cannot be defined using a Modula-2 definition, e.g. routines that have a variable number of parameters (see following example). Objects imported from module C have the following properties: - The object must be a procedure. - No type-checking is done on parameters to such procedures. - The procedures can receive any number of pararmeters. - The procedures can be called using either a procedure call or a function call. - Values returned by such procedures are of type WORD. Example: Give the IMPORT statement: FROM C IMPORT printf; all of the following statements are legal: printf; printf(string, integer, longreal); cardinal = printf(string, integer, longreal);
gak@mhuxm.UUCP (Vincent Hatem) (04/11/88)
In article <535@m10ux.UUCP>, rgr@m10ux.UUCP (Duke Robillard) writes: > In article <764@ndsuvax.UUCP> ncreed@ndsuvax.UUCP (Walter Reed) writes: > >Here is an example of why these things bother me. > > IF (x = TRUE) THEN > > WriteString("Error #"); > > WriteCard(e); > > WriteString(" in line #"); > > WriteCard(l); > > WriteString(" detected."); > : > : > the solution. Has anyone ever written some kind of front end for a > modula compiler to turn printf's into WriteSh*t's? It doesn't seem > : > | Duke Robillard > | AT&T Bell Labs m10ux!rgr@ihnp4.UUCP > | Murray Hill, NJ {any biggy}!ihnp4!m10ux!rgr Come on... give me a break. If you don't like the I/O subsystem, why don't you re-write it??? A front-end translator in m4 is silly. Besides, the whole idea of the WriteCard()-style I/O system is to remove the 20K overhead that printf() adds to programs. 95% of the functionality of printf() is never used, and you're just making your text segment larger. I've worked on a replacement for printf() in Modula-2 which allowed most of the basic formatting of numbers/strings which are used regularly. It also allowed for OOP Object I/O function definitions, which allow you to tell it to Write an Object with the functionality of a c++ construct such as this: cout << some_object_instance; I'd post it, but it is based on a new I/O system (which I also worked on), which is incompatible with Wirth's I/O system. (there used to be talk of making the library we wrote Freeware, but I haven't seen it yet...) -- Vincent Hatem AT&T International, International Systems Operations, UNIX Technical Support Telefon: (201) 953-8030 Please send all e-mail to: ihnp4!atti01!vch
rgr@m10ux.UUCP (Duke Robillard) (04/12/88)
In article <96@lzaz.ATT.COM> bds@lzaz.ATT.COM (BRUCE SZABLAK) writes: >In article <535@m10ux.UUCP>, rgr@m10ux.UUCP (Duke Robillard) writes: >> I hate C as much as the next guy, but it seems to me that printf is >> the solution. > >I like C++'s overloading of the << and >> operator's even better than >printf. It allows you to define custom print routines for each structure >(class) declared, and then to print (to stdout for example) you do: > cout << god_awful_structure_instance; Yeah, but that doesn't help me print out an error message, two strings, and an integer return code. I don't want to have to write a method for every print.... -- + | Duke Robillard | AT&T Bell Labs m10ux!rgr@ihnp4.UUCP | Murray Hill, NJ {any biggy}!ihnp4!m10ux!rgr
alan@pdn.UUCP (Alan Lovejoy) (04/12/88)
In article <96@lzaz.ATT.COM> bds@lzaz.ATT.COM (BRUCE SZABLAK) writes: >I like C++'s overloading of the << and >> operator's even better than >printf. It allows you to define custom print routines for each structure >(class) declared, and then to print (to stdout for example) you do: > cout << god_awful_structure_instance; >The advantage of this approach over using procedures in >modula2 or pascal is (besides previty) that you don't have to >remember the type of structure your variable is; the compiler figures it out. Overloading or procedure-name-abstraction is badly needed in Modula-2. However, it needs to be applicable to ANY AND ALL procedures and operators. For printing, I prefer: Print(object, fieldWidth); The programmer writes a Print procedure for each data type he defines and/or wants to print. Better yet, he writes a "BinToStr" procedure for each data type. Then Print can be defined generically once: PROCEDURE Print(object; fieldWidth: CARDINAL); VAR str: ARRAY [0..79] OF CHAR; BEGIN BinToStr(object, fieldWidth); Display.WriteString(str); END Print; Notice that this also requires the additional capability to define parameters of unspecified type. To compile this procedure, the compiler generates abstract pseudo-code which is data-type independent as its final output instead of object code. Object code for the procedure is generated each time the compiler discovers an actual call to the procedure, because only then can it know the data type of the typeless argument and generate the correct code. -- Alan Lovejoy; alan@pdn; 813-530-8241; Paradyne Corporation: Largo, Florida. Disclaimer: Do not confuse my views with the official views of Paradyne Corporation (regardless of how confusing those views may be). Motto: Never put off to run-time what you can do at compile-time!
bds@lzaz.ATT.COM (BRUCE SZABLAK) (04/13/88)
In article <547@m10ux.UUCP>, rgr@m10ux.UUCP (Duke Robillard) writes: > In article <96@lzaz.ATT.COM> bds@lzaz.ATT.COM (BRUCE SZABLAK) writes: > >I like C++'s overloading of the << and >> operator's even better ... > > Yeah, but that doesn't help me print out an error message, two strings, > and an integer return code. I don't want to have to write a method > for every print.... > In C++: #include <stream.h> Then: ... cout << "error message " << string1 << " " << string2 << " " << rc; ... The header file defines printing for the ususal types.
dhesi@bsu-cs.UUCP (Rahul Dhesi) (04/13/88)
In article <730@mhuxm.UUCP> gak@mhuxm.UUCP (Vincent Hatem) writes: >Besides, the whole idea of the WriteCard()-style I/O system is to remove the >20K overhead that printf() adds to programs. Actually, the story I hear is that WriteCard was originally meant (back when Wirth was developing languages on an old CDC machine) to punch a card. Nostalgia being what it is, WriteCard now prints unsigned integers. If you'll believe that, you'll also probably believe that Modula-2 is an improvement on Pascal. -- Rahul Dhesi UUCP: <backbones>!{iuvax,pur-ee,uunet}!bsu-cs!dhesi
mike@turing.UNM.EDU (Michael I. Bushnell) (04/13/88)
In article <730@mhuxm.UUCP> gak@mhuxm.UUCP (Vincent Hatem) writes: >Besides, the whole idea of the WriteCard()-style I/O system is to remove the >20K overhead that printf() adds to programs. 95% of the functionality of >printf() is never used, and you're just making your text segment larger. On 4.3 BSD on a Vax, I get the following: % size doprnt.o text data bss dec hex 2196 36 0 2232 8b8 % size printf.o text data bss dec hex 40 36 0 76 4c To use printf, it ups your text segment by a whole 2236 bytes. Wow. It also isn't too slow: 4.3BSD doprnt is coded in assembly language and manages to be quite fast. Puts is about 10% of that (200 bytes) but isn't really any faster. But CERTAINLY NOT a "20K" overhead. Go ahead and use printf. It doesn't really cost you that much. N u m q u a m G l o r i a D e o Michael I. Bushnell HASA - "A" division 14308 Skyline Rd NE Computer Science Dept. Albuquerque, NM 87123 OR Farris Engineering Ctr. OR University of New Mexico mike@turing.unm.edu Albuquerque, NM 87131 {ucbvax,gatech}!unmvax!turing.unm.edu!mike
faustus@ic.Berkeley.EDU (Wayne A. Christopher) (04/13/88)
In article <730@mhuxm.UUCP>, gak@mhuxm.UUCP (Vincent Hatem) writes: > Besides, the whole idea of the WriteCard()-style I/O system is to remove the > 20K overhead that printf() adds to programs. 95% of the functionality of > printf() is never used, and you're just making your text segment larger. I doubt this is the reason -- hardly anybody cares about text segment size any more, and in any case, printf is only 2K, not 20K. (In Ultrix -- your mileage may vary.) Wayne
uday@mips.COM (Robert Redford) (04/13/88)
In article <2608@bsu-cs.UUCP>, dhesi@bsu-cs.UUCP (Rahul Dhesi) writes: > Actually, the story I hear is that WriteCard was originally meant > (back when Wirth was developing languages on an old CDC machine) > to punch a card. Nostalgia being what it is, WriteCard now > prints unsigned integers. WriteCard stands for WriteCardinal. There is a standard library module CardinalIO that deals with the IO of cardinal numbers. In fact there is a predefined data type CARDINAL. It has nothing to do with cards. ..Uday
gjditchfield@violet.waterloo.edu (Glen Ditchfield) (04/13/88)
In article <547@m10ux.UUCP> rgr@m10ux.UUCP (Duke Robillard) writes: >In article <96@lzaz.ATT.COM> bds@lzaz.ATT.COM (BRUCE SZABLAK) writes: >>I like C++'s overloading of the << and >> operator's even better than >>printf. It allows you to define custom print routines for each structure >>(class) declared, and then to print (to stdout for example) you do: >> cout << god_awful_structure_instance; > >Yeah, but that doesn't help me print out an error message, two strings, >and an integer return code. I don't want to have to write a method >for every print.... Neither would I. Fortunately for us, the i/o operators can be chained. cout << "Error: file " << fname << ", line " << lnum << error_msg << "\n"; I think this is a big improvement over calls to overloaded "Print" procedures. Details: "<<" associates left-to-right. `cout << "Error: file "' returns cout as it's value, which is used as the left operand by `<< fname'. When streams like cout (or the standard input, "cin") are used in an integer context, the stream-to-int conversion returns the stream's status, so you can write input loops like while ( cin >> val ) { ... If a programmer-defined type must be read or written, the programmer writes "<<" and ">>" operators for it.
ecb@pluto.uucp (Eric Brown) (04/16/88)
In article <942@unmvax.unm.edu> mike@turing.UNM.EDU.UUCP (Michael I. Bushnell) writes: >To use printf, it ups your text segment by a whole 2236 bytes. Wow. >Puts is about 10% of that (200 bytes) >but isn't really any faster. But CERTAINLY NOT a "20K" overhead. Go >ahead and use printf. It doesn't really cost you that much. Now include the size of the floating point libraries. On Microsoft 5.0 C, calling printf links the floating point emulator, which is about 30K of code. Is this significant? I think so. I once hacked out 300K out of a suite of programs by transforming printf's to puts'es and also by using an integer-only printf. Sometimes, every byte counts. Not all the world is a vax, you know. -Eric, Called Ultrahacker
rgr@m10ux.UUCP (Duke Robillard) (04/17/88)
In article <6366@watdragon.waterloo.edu] gjditchfield@violet.waterloo.edu (Glen Ditchfield) writes: ]In article <547@m10ux.UUCP] rgr@m10ux.UUCP (Duke Robillard) writes: ]]In article <96@lzaz.ATT.COM] bds@lzaz.ATT.COM (BRUCE SZABLAK) writes: ]]] cout << god_awful_structure_instance; ]]Yeah, but that doesn't help me print out an error message, two strings, ]]and an integer return code. I don't want to have to write a method ]]for every print.... ] ]Neither would I. Fortunately for us, the i/o operators can be chained. ] cout << "Error: file " << fname << ", line " << lnum ] << error_msg << "\n"; This is cool. The syntax is a little bulky (how un-C-like) but overall it looks good. ]I think this is a big improvement over calls to overloaded "Print" ]procedures. However, I don't really see that this is any different than overloading a print procedure. all you're doing is overloading "<<" (i.e. making it handle strings, integers, reals, whatever), right? -- + | Duke Robillard | AT&T Bell Labs m10ux!rgr@ihnp4.UUCP | Murray Hill, NJ {any biggy}!ihnp4!m10ux!rgr
djones@megatest.UUCP (Dave Jones) (04/19/88)
in article <553@m10ux.UUCP>, rgr@m10ux.UUCP (Duke Robillard) says: ... [ cout << foo << bar; // etc. ] > > However, I don't really see that this is any different than > overloading a print procedure. all you're doing is overloading "<<" > (i.e. making it handle strings, integers, reals, whatever), right? > That's right. It's just easier to type and to read than having a jillion "print"'s. For better or worse (*better* in my opinion), C++ classes don't tend to "know how to print themselves". Instead you overload the << operator to print new classes. Also remember that in C++, "int" is not a class!! So the function can't be a member of class int. Translated into Smalltalk, the class "int" does not recognize the message "<<". "Print an int" _could_ be an external function, defined this way: ostream& operator<< (ostream& o, int a) {return o << (long)a;} There is an argument to be made that it SHOULD be declared this way, to reduce the number of member functions of ostream. That is to say, to reduce the number of types which ostream "knows about". But the way things are actually implemented, the function is a member of ostream, and has this declaration: ostream& operator<<(int a) { return *this<<long(a); }
chris@mimsy.UUCP (Chris Torek) (04/21/88)
>In article <942@unmvax.unm.edu> mike@turing.UNM.EDU.UUCP (Michael >I. Bushnell) writes: >>To use printf, it ups your text segment by a whole 2236 bytes. ... In article <230@jupiter.olyis.UUCP> ecb@pluto.uucp (Eric Brown) writes: >Now include the size of the floating point libraries. On Microsoft 5.0 C, >calling printf links the floating point emulator, which is about 30K of code. [stuff deleted] >Not all the world is a vax, you know. Conversely, not all the world is an IBM PC with a Microsoft compiler. It is not difficult[*] to arrange for the linker to pull in a simple printf() if the rest of the program does not use floating point, and a full-blown floating point printf if it does. Just because Microsoft's implementation is poor (in terms of space, at any rate) is no reason to castigate printf itself. ----- [*]What, never? Well, hardly ever. The trick is to have the linker examine the undefined externals for floating-point library references, and if present, link with `bigprintf.obj'; otherwise it should use `littleprintf.obj'. It is just a Small Matter of Programming. -- In-Real-Life: Chris Torek, Univ of MD Comp Sci Dept (+1 301 454 7163) Domain: chris@mimsy.umd.edu Path: uunet!mimsy!chris
dhesi@bsu-cs.UUCP (Rahul Dhesi) (04/22/88)
In article <11154@mimsy.UUCP> chris@mimsy.UUCP (Chris Torek) writes: >It is not difficult[*] to arrange for the linker to pull in a simple >printf() if the rest of the program does not use floating point, and a >full-blown floating point printf if it does. Turbo C 1.0 seems to do this. -- Rahul Dhesi UUCP: <backbones>!{iuvax,pur-ee,uunet}!bsu-cs!dhesi
friedl@vsi.UUCP (Stephen J. Friedl) (04/22/88)
In article <11154@mimsy.UUCP>, chris@mimsy.UUCP (Chris Torek) writes: > It is not difficult[*] to arrange for the linker to pull in a simple > printf() if the rest of the program does not use floating point, and a > full-blown floating point printf if it does. [....] > > [*]What, never? Well, hardly ever. The trick is to have the linker > examine the undefined externals for floating-point library references, > and if present, link with `bigprintf.obj'; otherwise it should use > `littleprintf.obj'. It is just a Small Matter of Programming. This brings up an interesting question. The C compiler on the Onyx Z8000 (R.I.P.) would generate a symbol "_fltused" if any floating point of any kind was used in that particular module. When the linker went to resolve everything, it loaded in the floating-point emulator if this symbol was found and didn't if it wasn't. As far as I could tell, this symbol was nothing more than a indicator to ld(1). Does anybody else do this? It seems so much more reasonable than the ubiquitous "-f" flag. -- Steve Friedl V-Systems, Inc. Resident access(2) basher friedl@vsi.com {backbones}!vsi.com!friedl attmail!vsi!friedl
henry@utzoo.uucp (Henry Spencer) (04/24/88)
> >Now include the size of the floating point libraries. On Microsoft 5.0 C, > >calling printf links the floating point emulator, which is about 30K of code. > >...Not all the world is a vax, you know. > > It is not difficult[*] to arrange for the linker to pull in a simple > printf() if the rest of the program does not use floating point, and a > full-blown floating point printf if it does... In fact, if your C implementation has a limited address space and a bulky floating-point library and does NOT do this, complain to the implementors. It's not a novel or strange idea. The very first C implementation, on the pdp11, used this trick. If Microslop can't be bothered, don't buy their compiler. -- "Noalias must go. This is | Henry Spencer @ U of Toronto Zoology non-negotiable." --DMR | {ihnp4,decvax,uunet!mnetor}!utzoo!henry
henry@utzoo.uucp (Henry Spencer) (04/24/88)
> ... The C compiler on the > Onyx Z8000 (R.I.P.) would generate a symbol "_fltused" if any > floating point of any kind was used in that particular module. > When the linker went to resolve everything, it loaded in the > floating-point emulator if this symbol was found and didn't if it > wasn't. As far as I could tell, this symbol was nothing more > than a indicator to ld(1). Does anybody else do this? ... Onyx borrowed this from the original pdp11 C implementation, in fact, which used this technique to pick up a non-floating-point printf if floating point was not in use. It didn't even require any special trickery in the linker: generate an external reference to fltused when floating-point is used; have two versions of the internal floating-point-formatting function, one which implements the full show and also defines fltused and one which does the integer subset and doesn't define fltused; and arrange for the fltused version to be seen by the linker before printf and the non-fltused version are seen. > ... It seems so much more reasonable than the ubiquitous "-f" flag. It's a reasonable substitute for -f only if you assume that the hardware never has proper floating-point support, or if you don't care about hauling the floating-point software along unnecessarily. The Onyx came under the former heading, as I recall. -- "Noalias must go. This is | Henry Spencer @ U of Toronto Zoology non-negotiable." --DMR | {ihnp4,decvax,uunet!mnetor}!utzoo!henry
gwyn@brl-smoke.ARPA (Doug Gwyn ) (04/24/88)
In article <230@jupiter.olyis.UUCP> ecb@pluto.UUCP (Eric Brown) writes: >In article <942@unmvax.unm.edu> mike@turing.UNM.EDU.UUCP (Michael I. Bushnell) writes: >>Go ahead and use printf. It doesn't really cost you that much. >Now include the size of the floating point libraries. Sounds to me like to need to have a little talk with your compiler vendor. There is no need to link floating-point support into the process image when the application has made no use of floating point. Even the old PDP-11 UNIX got this right -- there were two _doprnt entries in the library, one for use with floating-point and one without; the choice was tied to whether or not the compiler had generated a ".fltused" reference.
mcdonald@uxe.cso.uiuc.edu (04/25/88)
>In fact, if your C implementation has a limited address space and a bulky >floating-point library and does NOT do this, complain to the implementors. >It's not a novel or strange idea. The very first C implementation, on the >pdp11, used this trick. If Microslop can't be bothered, don't buy their >compiler. -- The following program #include <stdio.h> main() { int i = 5; printf("%d is an int\n",i); } compiles and links under Microsoft C 5.00 to a .exe file 7089 bytes long. Changing "int" to "double" and %d to %lf results in a file 21242 bytes long. The Microsoft C compiler is quite good, better than most Unix ones, I would guess. It has myriads of options that allow you do get just what you want. Doug McDonald
swarbric@tramp.Colorado.EDU (Frank Swarbrick) (04/27/88)
In article <225800026@uxe.cso.uiuc.edu> mcdonald@uxe.cso.uiuc.edu writes:
:The following program
:#include <stdio.h>
:main()
:{
: int i = 5;
: printf("%d is an int\n",i);
:}
:compiles and links under Microsoft C 5.00 to a .exe file 7089 bytes long.
:Changing "int" to "double" and %d to %lf results in a file 21242 bytes long.
:The Microsoft C compiler is quite good, better than most Unix ones, I would
:guess. It has myriads of options that allow you do get just what you want.
I just thought I'd mention this, Turbo C 1.5 gets EXE files of 5750 bytes
and 19662 bytes respectivly with this code...
Frank Swarbrick (and his cat) swarbric@tramp.Colorado.EDU
...{ncar!nbires}!boulder!tramp!swarbric
"The shifting shafts of shining weave the fabric of their dreams..."