As 64 bits support is not expected in the next version it is no longer an option to wait for the possibility to migrate our existing code base to unicode and 64-bit in one go.
However it would be nice if we could already prepare our code for 64-bit when doing our unicode translation. This will minimize impact in the event it will finally appear in version 2020.
Any suggestions how to approach this without introducing to much clutter if it doesn't arrive until 2020?
There's another similar question, but I'll repeat my reply here too, to make sure as many people see this info:
First up, a disclaimer: although I work for Embarcadero. I can't speak for my employer. What I'm about to write is based on my own opinion of how a hypothetical 64-bit Delphi should work, but there may or may not be competing opinions and other foreseen or unforeseen incompatibilities and events that cause alternative design decisions to be made.
That said:
There are two integer types, NativeInt and NativeUInt, whose size will
float between 32-bit and 64-bit depending on platform. They've been
around for quite a few releases. No other integer types will change size
depending on bitness of the target.
Make sure that any place that relies on casting a pointer value to an
integer or vice versa is using NativeInt or NativeUInt for the integer
type. TComponent.Tag should be NativeInt in later versions of Delphi.
I'd suggest don't use NativeInt or NativeUInt for non-pointer-based values. Try to keep your code semantically the same between 32-bit and 64-bit. If you need 32 bits of range, use Integer; if you need 64 bits, use Int64. That way your code should run the same on both bitnesses. Only if you're casting to and from a Pointer value of some kind, such as a reference or a THandle, should you use NativeInt.
Pointer-like things should follow similar rules to pointers: object
references (obviously), but also things like HWND, THandle, etc.
Don't rely on internal details of strings and dynamic arrays, like
their header data.
Our general policy on API changes for 64-bit should be to keep the
same API between 32-bit and 64-bit where possible, even if it means that
the 64-bit API does not necessarily take advantage of the machine. For
example, TList will probably only handle MaxInt div SizeOf(Pointer)
elements, in order to keep Count, indexes etc. as Integer. Because the
Integer type won't float (i.e. change size depending on bitness), we
don't want to have ripple effects on customer code: any indexes that
round-tripped through an Integer-typed variable, or for-loop index,
would be truncated and potentially cause subtle bugs.
Where APIs are extended for 64-bit, they will most likely be done with
an extra function / method / property to access the extra data, and this
API will also be supported in 32-bit. For example, the Length() standard
routine will probably return values of type Integer for arguments of
type string or dynamic array; if one wants to deal with very large
dynamic arrays, there may be a LongLength() routine as well, whose
implementation in 32-bit is the same as Length(). Length() would throw
an exception in 64-bit if applied to a dynamic array with more than 232
elements.
Related to this, there will probably be improved error checking for
narrowing operations in the language, especially narrowing 64-bit values
to 32-bit locations. This would hit the usability of assigning the
return value of Length to locations of type Integer if Length(),
returned Int64. On the other hand, specifically for compiler-magic
functions like Length(), there may be some advantage of the magic taken,
to e.g. switch the return type based on context. But advantage can't be
similarly taken in non-magic APIs.
Dynamic arrays will probably support 64-bit indexing. Note that Java
arrays are limited to 32-bit indexing, even on 64-bit platforms.
Strings probably will be limited to 32-bit indexing. We have a hard
time coming up with realistic reasons for people wanting 4GB+ strings
that really are strings, and not just managed blobs of data, for which
dynamic arrays may serve just as well.
Perhaps a built-in assembler, but with restrictions, like not being able to freely mix with Delphi code; there are also rules around exceptions and stack frame layout that need to be followed on x64.
First, look at the places where you interact with non-delphi libraries and api-calls,
they might differ. On Win32, libraries with the stdcall calling convenstion are named like _SomeFunction#4 (#4 indicating the size of the parameters, etc). On Win64, there is only one calling convention, and the functions in a dll are no more decorated. If you import functions from dll files, you might need to adjust them.
Keep in mind, in a 64 bit exe you cannot load a 32-bit dll, so, if you depend on 3rd party dll files, you should check for a 64-bit version of those files as well.
Also, look at Integers, if you depend on their max value, for example when you let them overflow and wait for the moment that happens, it will cause trouble if the size of an integer is changed.
Also, when working with streams, and you want to serialize different data, with includes an integer, it will cause trouble, since the size of the integer changed, and your stream will be out of sync.
So, on places where you depend on the size of an integer or pointer, you will need to make adjustments. When serializing sush data, you need to keep in mind this size issue as well, as it might cause data incompatibilities between 32 and 64 bit versions.
Also, the FreePascal compiler with the Lazarus IDE already supports 64-bit. This alternative Object Pascal compiler is not 100% compatible with the Borland/Codegear/Embarcadero dialect of Pascal, so just recompiling with it for 64-bit might not be that simple, but it might help point out problems with 64-bit.
The conversion to 64bit should not be very painful. Start with being intentional about the size of an integer where it matters. Don't use "integer" instead use Int32 for integers sized at 32bits, and Int64 for integers sized at 64bits. In the last bit conversion the definition of Integer went from Int16 to Int32, so your playing it safe by specifying the exact bit depth.
If you have any inline assembly, create a pascal equivalent and create some unit tests to insure that they operate the same way. Perform some timing tests of both and see if the assembly still runs faster enough to keep. If it does, then you will want to make changes to both as they are needed.
Use NativeInt for integers that can contain casted pointers.
Related
A u32 takes 4 bytes of memory, a String takes 3 pointer-sized integers (for location, size, and reserved space) on the stack, plus some amount on the heap.
This to me implies that Rust doesn't know, when the code is executed, what type is stored at a particular location, because that knowledge would require more memory.
But at the same time, does it not need to know what type is stored at 0xfa3d2f10, in order to be able to interpret the bytes at that location? For example, to know that the next bytes form the spec of a String on the heap?
How does Rust store types at runtime?
It doesn't, generally.
Rust doesn't know, when the code is executed, what type is stored at a particular location
Correct.
does it not need to know what type is stored
No, the bytes in memory should be correct, and the rest of the code assumes as much. The offsets of fields in a struct are baked-in to the generated machine code.
When does Rust store something like type information?
When performing dynamic dispatch, a fat pointer is used. This is composed of a pointer to the data and a pointer to a vtable, a collection of functions that make up the interface in question. The vtable could be considered a representation of the type, but it doesn't have a lot of the information that you might think goes into "a type" (unless the trait requires it). Dynamic dispatch isn't super common in Rust as most people prefer static dispatch when it's possible, but both techniques have their benefits.
There's also concepts like TypeId, which can represent one specific type, but only of a subset of types. It also doesn't provide much capability besides "are these the same type or not".
Isn't this all terribly brittle?
Yes, it can be, which is one of the things that makes Rust so interesting.
In a language like C or C++, there's not much that safeguards the programmer from making dumb mistakes that go out and mess up those bytes floating around in memory. Making those mistakes is what leads to bugs due to memory safety. Instead of interpreting your password as a password, it's interpreted as your username and printed out to an attacker (oops!)
Rust provides safeguards against that in the form of a strong type system and tools like the borrow checker, but still all done at compile time. Unsafe Rust enables these dangerous tools with the tradeoff that the programmer is now expected to uphold all the guarantees themselves, much like if they were writing C or C++ again.
See also:
When does type binding happen in Rust?
How does Rust implement reflection?
How do I print the type of a variable in Rust?
How to introspect all available methods and members of a Rust type?
I run Lua on a CPU without dedicated floating point HW, depending on SW emulation.
From luaopt.h I can see that some macros are set to double, but it does not clearly state when floats are used and its a little hard to track it.
If my script does simple stuff like:
a=0
a=a+1
for...
Would that involve a floating point operations at any level?
If no that's fine, but what is then the benefit to change macros to long?
(I tried of course but did not work....)
All numeric operations in Lua are performed (according to the default configuration) in floating point. There is no distinction made between floating point and integer, all values are simply numbers.
The actual C type used to store a Lua number is set in luaconf.h, and it is both allowed and even practical to change that to a suitable integral type. You start by changing LUA_NUMBER from double to int, long, or perhaps ptrdiff_t. Then you will find you need to tweak the related macros that control the conversions between strings and numbers. And, of course, you will likely need to eliminate most or all of the base math library since math.sin() and its friends and neighbors are not particularly useful over integers.
The result will be a Lua interpreter where all numbers are integers. The language will still allow you to type 3.14, but it will be stored as 3. Your code will likely not be completely portable to a Lua interpreter built with the standard configuration since a huge amount of Lua code casually assumes that floating point arithmetic is permitted, and remember that your compiled byte code will definitely not be compatible since byte code will store numbers as LUA_NUMBER.
There is LNUM patch (used, for example, by OpenWrt project which relies heavily on Lua for providing Web UI on hardware without FPU) that allows dual integer/floating point representation of numbers in Lua with conversions happening behind the scenes when required. With it most integer computations will be performed without resorting to FPU. Unfortunately, it's only applicable to Lua 5.1; 5.2 is not supported.
In Delphi, why are AnsiStrings indexed from one and dynamic arrays indexed from zero? Is this a historical accident, to make AnsiStrings work more like ShortStrings, or is there some deeper logic at work?
One of the contributing factors that led to "Pascal" strings being 1 indexed instead of 0 indexed was that the length of the string was stored in the zeroth byte. Yes, that could have been hidden from the programmer's view by having the compiler internally add a constant offset to the string index expression (as was done in Delphi's long strings later) but in the beginning things were much simpler. Allocate a block of memory, store the length in byte zero, index the char data from byte 1. End of story.
As I recall UCSD Pascal was using this length-in-zero-byte convention long before Turbo Pascal came along.
As for why dynamic arrays are zero based, I don't recall any specific reason but I would guess it reflects the dynamic array's kinship to dynamically allocating a buffer and indexing off the buffer pointer. The array types that you would use to create array pointer types were zero based arrays. The first byte is found at buffer pointer + 0 offset. This is the C rationalization for zero based everything. There was no compelling reason to carry string's 1 based indexing pattern over to compiler managed arrays when string's 1 based indexing was already (and had always been) the exception rather than the norm.
It may well be that because the string type was the first array-like data type that everyone first encountered and possibly the most used data type across the board, there may be a perception of a bias towards 1 based indexing in the language. However, if you look closely I think you'll find arrays in Pascal (distinct from string) have never been inherently 1 based, especially when dynamically allocated.
The reason for the Delphi string tradition of 1-based strings is quite simple. The tradition comes from the implementation of old style Turbo Pascal strings. That data type stored the length of the string in the first byte of the variable, index 0. The string data began in the next byte, index 1.
You can still use that data type today. It's now called ShortString. As is immediately obvious from it's implementation, there is a 255 character limit. This limit led to the introduction of huge strings, if I recall correctly, in Delphi 2. When huge strings were introduced the language designers chose to retain 1-based indexing to make it easier for developers to switch from short strings to huge strings.
I guess Turbo Pascal didn't invent the idea of using element 0 for length. It's just that I'm too young to remember what came before then!
Dynamic arrays weren't bound by the past in the same way and had a free choice. I don't know why zero based was chosen. Perhaps because it fits more easily with the prevailing fashion on platform on which Delphi existed at that time, namely Windows. That's just a guess though. Danny Thorpe worked on the Delphi compiler at that time, and even he can't remember the rationale!
The Delphi language designers are currently moving towards zero based string indexing for huge strings. The initial steps in this direction can be seen in XE3 in the TStringHelper class which uses 0-based indexing. And also in the ZEROBASEDSTRINGS conditional which allows you to opt in to 0-based indexing. Expect the next generation Delphi compiler to use 0-based indexing only. The times they are changin'.
Historical accident.
Pascal strings and arrays traditionally start at 1.
C - and perhaps consequently AnsiStrings - start at 0.
I don't know the rationale for "breaking with Pascal tradition" for Dynamic arrays, which also start at zero. But it makes sense, and I agree with it ...
IMHO...
I'm not a Pascal newbie, but I still don't know until now why Delphi and Free Pascal usually declares parameters and returned values as signed integers whereas I see them should always be positive. For example:
Pos() returns type of Integer. Is it possible to be a negative?
SetLength() declares the NewLength parameter as a type of Integer. Is there a negative length for string?
System.THandle declared as Longint. Is there a negative number for handles?
There are many decisions like those in Delphi and Free Pascal. What considerations were behind this?
In Pascal, Integer (signed) is the base type. All other integer number types are a subranges of integer. (this is not entirely true in Borland dialects, given longint in TP and int64 in Delphi, but close enough).
An important reason for that if the intermediate result of calculations gets negative, and you calculate with unsigned integers, range check errors will trigger, and since most older programming languages DON'T assume 2-complement integers, the result (with range checks off) might even be corrupt.
The THandle case is much simpler. Delphi didn't have a proper 32-bit unsigned till D4, but only a 31-bit cardinal. (since 32-bit unsigned integer is not a subrange of integer, the later unsigned ints are a subset of int64, which moved the problem to uint64 which was only added in D2010 or so)
So in many places in the headers signed types are used where the winapi uses unsigned types, probably to avoid the 32th bit getting accidentally corrupt in those versions, and the custom stuck.
But the winapi case is different from the general case.
Added later Some Pascal (and Modula2/3) implementations circumvent this trap by setting the integer at a size larger than the wordsize, and require all numeric types to declare a proper subrange, like in the below program.
The first holds the primary assumption that everything is a subset of integer, and the second allows the compiler to scale nearly everything down again to fit in registers, specially if the CPU has some operations for larger than word operations. (like x86 where 32-bit * 32-bit mul gives a 64-bit result, or can detect wordsize overflows using status bits (e.g. to generate range exceptions for adds without doing a full 2*wordsize add)
var x : 0..20;
y : -10..10;
begin
// any expression of x and y has a range -10..20
Turbo Pascal and Delphi emulate an integer type twice the wordsize for their 16-bit and 32-bit offerings. The handling of the highest unsigned type is hacky at best.
Well, for a start THandle is declared incorrectly. It's unsigned in the Windows headers and should be so in Delphi. In fact I think this was corrected in a recent release of Delphi.
I'd imagine that the preference for signed over unsigned is largely historical and not particularly significant. However, I can think of one example where it is important. Consider the for loop:
for i := 0 to Count-1 do
If i is unsigned and Count is 0 then this loop runs from 0 to $FFFFFFFF which is not what you want. Using a signed integer loop variable avoids that problem.
Pascal is a victim of its syntax here. The equivalent C or C++ loop has no such trouble
for (unsigned int i=0; i<Count; i++)
due to the syntactic difference and use of a comparison operator as stopping condition.
This could also be the reason why Length() on a string or dynamic array returns a signed value. And so for consistency, SetLength() should accept signed values. And given that the return value of Pos() is used to index strings, it should be signed also.
Here's another Stack Overflow discussion of the topic: Should I use unsigned integers for counting members?
Of course, I'm speculating wildly here. Perhaps there was no design and just out of habit the precedent of using signed values was set and became enshrined.
Some string related search functions return -1 when nothing is found.
I believe the reasoning behind this is that MaxInt is 2GB which is the maximum size for strings in 32 bit Delphi. This because a single process can have up to 2GB memory
There are many reasons for using signed integers, even some that might apply when you do not intend to return a negative value.
Imagine I write code that calls Pos, and I want to do math with the results. Would you rather have a negative result (Pos('x',s)-5) raise a range-check exception, underflow and become a very large unsigned number around 4 billion, or go negative, if Pos('x',s) returns 1? Either one is a source of problems for new users who seldom think about these cases, but the long-established tradition is that by using Integer results, it's your job to check for negative and zero results and not use them as string offsets. There is an advantage for beginning and for advanced programmers, in using Integer, and not having "negative" values roll under and become large unsigned values or raise range exceptions.
Secondly, remember that in beginning programming, one usually introduces Integer (signed) types long before one introduces unsigned types like Cardinal. Beginners often work with functions like Pos, and it makes sense to use the type that will create the least-unfriendly set of side effects. There are no negative side effects to having a range larger than the one you absolutely need (the range you probably need for Pos is 1 to maximum-string-length-in-delphi). There is zero benefit in 32-bit Delphi to using the Cardinal type for Pos, and there definitely ARE downsides to choosing it.
Once you get to 64-bit delphi, however, you could theoretically have strings LARGER than an Integer can hold, and moving to Cardinal wouldn't fix all your potential problems. However, the chance of anyone having a 2+ GB string is probably nil, and Delphi 64-bit compiler doesn't allow a >2 GB string, anyway. In my testing, I can achieve an almost 1 GB String in 64 bit Delphi. So the practical length limit for a Win64 string is about a billion (1073741814) characters, which is using nearly 2 GB of actual RAM. At that limit, I either get EIntOverflow or EAccessViolation, and it seems I am hitting Delphi run time library (RTL) bugs, not properly defined limits, so your mileage may vary.
I finally upgraded to Delphi XE. I have a library of units where I use strings to store plain ANSI characters (chars between A and U). I am 101% sure that I will never ever use UNICODE characters in those places.
I want to convert all other libraries to Unicode, but for this specific library I think it will be better to stick with ANSI. The advantage is the memory requirement as in some cases I load very large TXT files (containing ONLY Ansi characters). The disadvantage might be that I have to do lots and lots of typecasts when I make those libraries to interact with normal (unicode) libraries.
There are some general guidelines to show when is good to convert to Unicode and when to stick with Ansi?
The problem with general guidelines is that something like this can be very specific to a person's situation. Your example here is one of those.
However, for people Googling and arriving here, some general guidelines are:
Yes, convert to Unicode. Don't try to keep an old app fully using AnsiStrings. The reason is that the whole VCL is Unicode, and you shouldn't try to mix the two, because you will convert every time you assign a Unicode string to an ANSI string, and that is a lossy conversion. Trying to keep the old way because it's less work (or some similar reason) will cause you pain; just embrace the new string type, convert, and go with it.
Instead of randomly mixing the two, explicitly perform any conversions you need to, once - for example, if you're loading data from an old version of your program you know it will be ANSI, so read it into a Unicode string there, and that's it. Ever after, it will be Unicode.
You should not need to change the type of your string variables - string pre-D2009 is ANSI, and in D2009 and alter is Unicode. Instead, follow compiler warnings and watch which string methods you use - some still take an AnsiString parameter and I find it all confusing. The compiler will tell you.
If you use strings to hold bytes (in other words, using them as an array of bytes because a character was a byte) switch to TBytes.
You may encounter specific problems for things like encryption (strings are no longer byte/characters, so 'character' for 'character' you may get different output); reading text files (use the stream classes and TEncoding); and, frankly, miscellaneous stuff. Search here on SO, most things have been asked before.
Commenters, please add more suggestions... I mostly use C++Builder, not Delphi, and there are probably quite a few specific things for Delphi I don't know about.
Now for your specific question: should you convert this library?
If:
The values between A and U are truly only ever in this range, and
These values represent characters (A really is A, not byte value 65 - if so, use TBytes), and
You load large text files and memory is a problem
then not converting to Unicode, and instead switching your strings to AnsiStrings, makes sense.
Be aware that:
There is an overhead every time you convert from ANSI to Unicode
You could use UTF8String, which is a specific type of AnsiString that will not be lossy when converted, and will still store most text (Roman characters) in a single byte
Changing all the instances of string to AnsiString could be a bit of work, and you will need to check all the methods called with them to see if too many implicit conversions are being performed (for performance), etc
You may need to change the outer layer of your library to use Unicode so that conversion code or ANSI/Unicode compiler warnings are not visible to users of your library
If you convert to Unicode, sets of characters (can't remember the syntax, maybe if 'S' in MySet?) won't work. From your description of characters A to U, I could guess you would like to use this syntax.
My recommendation? Personally, the only reason I would do this from the information you've given is the memory use, and possibly performance depending on what you're doing with this huge amount of A..Us. If that truly is significant, it's both the driver and the constraint, and you should convert to ANSI.
You should be able to wrap up the conversion at the interface between this unit and its clients. Use AnsiString internally and string everywhere else and you should be fine.
In general only use AnsiString if it is important that the Chars are single bytes, Otherwise the use of string ensures future compatibility with Unicode.
You need to check all libraries anyway because all Windows API functions in Delhpi XE replaced by their unicode-analogues, etc. If you will never use UNICODE you need to use Delphi 7.
Use AnsiString explicitly everywhere in this unit and then you'll get compiler warning errors (which you should never ignore) for String to AnsiString conversion errors if you happen to access the routines incorrectly.
Alternately, perhaps preferably depending on your situation, simply convert everything to UTF8.
Stick with Ansi strings ONLY if you do not have the time to convert the code properly. The use of Ansi strings is really only for backward compatibility - to my knowledge C# does not have an equiavalent to Ansi strings. Otherwise use the standard Unicode strings. If you have a look on my web-site I have a whole strings routines unit (about 5,000 LOC) that works with both Delphi 2007 (non-Uniocde) and XE (Unicode) with only "string" interfaces and contains almost all of the conversion issues you might face.