scjones@sdrc.UUCP (Larry Jones) (10/15/88)
I am looking for information about C compilers that are overzealous about aligning structure members and padding structures. By this I mean that the compiler aligns more strictly than is required by the underlying hardware - for example, aligning an array of char on an int boundary rather than a char boundary. Similarly, a structure may be padded out to an even number of ints even though all the members are only char. Although this behavior is acceptable under the draft ANSI standard, it makes it impossible to declare a struct that conforms to some external format (e.g. records in a file). If a compiler does this by default but has some way to prevent it, please let me know what compiler and how to disable it (e.g. compiler switch, pragma, etc.). If a compiler doesn't have a way to disable it, I DEFINITELY what to know! If there are lots of compilers like this (which I doubt since I've just run into my first and I've dealt with LOTS before), then we're going to have to do a major rethink of a product we were just about to announce. If there are only a few (fingers crossed), we can just write them off. If you know of a compiler like this, please mail me the info and I will summarize. ---- Larry Jones UUCP: uunet!sdrc!scjones SDRC scjones@sdrc.uucp 2000 Eastman Dr. BIX: ltl Milford, OH 45150 AT&T: (513) 576-2070 "Save the Quayles" - Mark Russell
henry@utzoo.uucp (Henry Spencer) (10/16/88)
In article <410@sdrc.UUCP> scjones@sdrc.UUCP (Larry Jones) writes: >I am looking for information about C compilers that are overzealous >about aligning structure members and padding structures... >If there are lots of compilers like this ... >then we're going to have to do a major rethink... You're going to have to do a major rethink. Such compilers are indeed common. Many compilers pad all structs out to a word boundary just on general principles. Some of them actually have important reasons for doing so: on any machine where char pointers are a hassle -- Data General and Cray, to name two -- there is a strong efficiency incentive to pad all structs out so they can use word pointers. Life is also much simpler if all struct pointers share the same representation, which again bites you if char pointers are different. >... it makes it impossible to declare a struct that >conforms to some external format (e.g. records in a file)... Right. This can't be done portably in C. You have to read the data in as a lump of bytes and then do a component-by-component copy. (This is necessary anyway for ints and such, due to byte-ordering problems. To say nothing of questions like the size of ints and the representation of floating point...) Yes, this is a hassle. -- The meek can have the Earth; | Henry Spencer at U of Toronto Zoology the rest of us have other plans.|uunet!attcan!utzoo!henry henry@zoo.toronto.edu
rkl1@hound.UUCP (K.LAUX) (10/17/88)
The best way is to align the structure yourself! Sounds as if you have structures where the first field is a 'char', so you don't know how the compiler will align it. So don't make the first field a char...maybe you might have to insert a dummy 'int' field if all you have are 'char' fields. Alternatively, you could use a *union* of an 'int' with your structure of 'char' fields. What you're doing is allowing the *compiler* to decide how to align instead of *telling* the compiler how to align... --rkl
scjones@sdrc.UUCP (Larry Jones) (10/18/88)
In article <410@sdrc.UUCP>, I wrote: > I am looking for information about C compilers that are overzealous > about aligning structure members and padding structures. By this > I mean that the compiler aligns more strictly than is required by > the underlying hardware - for example, aligning an array of char > on an int boundary rather than a char boundary. Similarly, a > structure may be padded out to an even number of ints even though > all the members are only char. Well, I've been inundated by the first wave of replies saying "LOTS of compilers do this!" which means I didn't make myself clear and, sure enough, as I reread what I said, not only did I not make myself clear, I didn't say what I meant at all! What I'm interested in knowing about is NOT compilers that align more strictly than the hardware requires, but rather compilers that align more strictly in some cases than in others. For example, if a compiler chooses to align char on a multiple of 1, short on a multiple of 2, and long on a multiple of 4, that's OK by me. If, however, it further aligns char arrays or char structure members on a multiple of 2 or 4 (even though char is only aligned on a multiple of 1), THAT'S what I want to know about (particularly if there's no way to disable it). So far, the only compilers I know of that behave this way is the Microsoft V4 (and related Xenix) compiler (which had a switch to disable it), and the Apollo compiler (which does not). ---- Larry Jones UUCP: uunet!sdrc!scjones SDRC scjones@sdrc.uucp 2000 Eastman Dr. BIX: ltl Milford, OH 45150 AT&T: (513) 576-2070 "Save the Quayles" - Mark Russell
scs@athena.mit.edu (Steve Summit) (10/23/88)
In article <410@sdrc.UUCP> scjones@sdrc.UUCP (Larry Jones) writes: >I am looking for information about C compilers that are overzealous >about aligning structure members and padding structures... >it makes it impossible to declare a struct that >conforms to some external format (e.g. records in a file) As was discussed recently in this newsgroup, you can avoid all structure arrangement problems by simply not attempting to conform to external binary formats, but by using an external text file format instead. (Efficiency hackers find this solution unpalatable, but the parse time and file size issues are often not real problems in practice.) Steve Summit scs@adam.pika.mit.edu
cd@hrc63.co.uk (Colin Denman "GECCL") (11/01/88)
In article <7613@bloom-beacon.MIT.EDU>, scs@athena.mit.edu (Steve Summit) writes: > In article <410@sdrc.UUCP> scjones@sdrc.UUCP (Larry Jones) writes: > >[...] > > As was discussed recently in this newsgroup, you can avoid all ^^^ > structure arrangement problems by simply not attempting to > conform to external binary formats, but by using an external text > file format instead. (Efficiency hackers find this solution > unpalatable, but the parse time and file size issues are often > not real problems in practice.) My most frequent problems with enforced alignments occur with *hardware* formats and ones over which I have no control. What external text format do I use for them :->. If my machine doesn't have floating point hardware, I still expect a worthy C implementation to simulate with loss of efficiency. Why can't the same attitude be adopted to matters of alignments and bit-manipulations. It would be neat and portable if I could express a format (bitfields and all) secure in the knowledge that the only thing I loose is efficiency. If my solution permits, I can optimise data layouts, but at least I get something that *works*. It is not just a problem for "efficiency hackers", quite the opposite. Colin J Denman