An algo for making messy random strings.

[Steve Hutchesson]

The algo is to produce random string data to be used with Windows Class names or the strings for private messages. It is primarily to make this data impossible to predict as it changes every time the app is started.

http://masm32.com/board/index.php?topic=8881.0

[Ribeiro Alvo]

[Steve Hutchesson]

[David Roberts]

Another way:

#Compile exe
#Dim All
#Include "NTSecAPI.inc"

FunctionPBmain() As Long
Local szClassName as String * 64
RtlGenRandom( szClassName, 64 )
Waitkey$
End Function

[Stuart McLachlan]

Another way:

There must be dozens of ways at least.
Creating a random string is about week 2 of "Programming 101". Does it really warrant an entry in the Source Code Library"?

[David Roberts]

Does it really warrant an entry in the Source Code Library"?

I do not think that mine does which was posted because I felt that Steve's method was a bit of a 'dog's dinner'. He seems to be flogging the death out of ! rdtsc, ! bswap eax.

Oops, this is not the place for this conversation.

[Stuart McLachlan]

So he's still using the totally pointless bswap ?

[Steve Hutchesson]

The level of literacy here is truly mind boggling.

I flog to death MOV ADD SUB CMP JMP Jxx MOVSB MOVSD MOVSQ and a whole host of other Intel mnemonics, what is the bias against RDTSC BSWAP and MOV ?

It does what it does and while it appears that Stuart can live with sequential number sequences that are easily predicted, many of us can't when we need very hard (read impossible) to predict seed numbers.

> RtlGenRandom( szClassName, 64 )

I imagine it would work but why call a random number generator as a seeding source when you can do it with two instructions ? Surely code economy (read small and fast) has some mileage here.

You guys must be bored, now here is a suggestion, take up assembler programming, the easy stuff in 32 bit or the PHUN stuff in 64 bit.

Brought to you from my magnificent 12 core Xeon which I have just about finished, hardware is up and going, just added another 4tb disk and all of the software is close to configured.

[Steve Hutchesson]

Just so that you don't feel that there is enough variation in the simple reseed technique, here is a slightly more complex version of a reseeding algo in 64 bit.

; ¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤

 NOSTACKFRAME

 reseed proc

    mov rax, 100                                ; out of range number
    cpuid                                       ; set all regs to 0
    rdtsc                                       ; date time counter
    pause                                       ; spinlock pause
    bswap rax                                   ; reverse byte order
    mov r11, rax                                ; store rax in r11

    mov rcx, 10                                 ; loop count
  @@:
    rdtsc                                       ; date time counter
    pause                                       ; spinlock pause
    bswap rax                                   ; reverse byte order
    rol rax, 7                                  ; rotate left by prime
    xor r11, rax                                ; xor rax to r11
    rol r11, 5                                  ; rotate left by prime
    sub rcx, 1                                  ; decrement counter

    jnz @B
    mov rax, r11                                ; return the value in r11
    ret

 reseed endp

 STACKFRAME

; ¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤

[David Roberts]

The level of literacy here is truly mind boggling.

Thanks, Steve, I feel sure that Stuart will appreciate the compliment as well.

what is the bias against RDTSC BSWAP and MOV ?

What bias? I cannot be bothered to search for it but you introduced the '! rdtsc, ! bswap eax' combo quite a while ago and reintroduced it again only recently. You then use it again in the opening post's link. That is what I meant with "flogging the death out of ...".

In one of Melissa O'Neill's blogs she looks at seeding, and she cites, among others, using rdtsc. Using bswap made sense to me, destroying the sequential aspect of rdtsc, and I could not understand the objections to it.

but why call a random number generator as a seeding source

That is not unusual: Intel's RdSeed was developed to do just that; many folk use the KISS generator to seed Mersenne Twister. My modus operandi is to use a CPRNG to seed a PRNG.

when you can do it with two instructions ?

You are using two instructions with the core idea but you are using quite a few more in your link.

Surely code economy (read small and fast) has some mileage here.

I don't think that 'RtlGenRandom( szClassName, 64 )' will leave you short of RAM and, on my machine, takes 13.5µs to execute so it is not exactly dragging its feet.

now here is a suggestion, take up assembler programming

Funny you should suggest that because forty years ago, next year, I did just that using machine code on the ZX81. I say machine code because at the time I did not have an assembler and typed in hex code. Eventually, I got Hisoft Devpac. More than seventy-five per cent of my coding on the Atari ST was written in assembler. Admittedly, I have not done a great deal with Intel stuff at PowerBASIC and I do even less at FreeBASIC because the binaries produced by the C optimizing compiler are so fast. So don't assume from your ivory tower that the rest of haven't a clue about assembler programming.

Brought to you from my magnificent 12 core Xeon which I have just about finished, hardware is up and going, just added another 4tb disk and all of the software is close to configured.

What does that have to do with the price of bacon?

Am I being over critical? I don't think so as I reckon your post #8 is not up to your usual standards.

However, I like the look of your post #9.

FreeBASIC has an inline 64-bit assembler and knows what a 64-bit unsigned integer is, so I could put rax into a function unsigned __int64 return value.

[David Roberts]

Steve, question: What is 'mov rax, 100 ; out of range number' for?

[Dale Yarker]

RAX - 64 bit register A
EAX - 32 bit register A (low half of RAX)
AX - 16 bit register A (low 16 bits of EAX)
AL - low 8 bit register (low half of AX)
AH - high 8 bit register (high half of AX)
same pattern for B, C and D
see diagram above https://en.wikipedia.org/wiki/X86#64-bit part of page. There are more registers.

(David, maybe others looking too, so I put more than you probably need.)

((or did you mean the out of range part??? ))

[Stuart McLachlan]

Using bswap made sense to me, destroying the sequential aspect of rdtsc, and I could not understand the objections to it.

WIth a register that is rotating through the 2^32 values in the DWORD about every 2 seconds and a single request for a seed for the RNG, what is the "sequential" aspect you are so concerned about?

Unless you are re-seeding the RNG repeatedly in a very short time frame there is no "sequential aspect".. The BSWAP is superfluous in this use case

[Steve Hutchesson]

> What bias? I cannot be bothered to search for it but you introduced the '! rdtsc, ! bswap eax' combo quite a while ago and reintroduced it again only recently. You then use it again in the opening post's link. That is what I meant with "flogging the death out of ...".

You appear to have missed what the posted algo does, it seeded the PB randomize so that the following loop could produce a sequence of random characters. It used the old PB rnd so it was easier to understand and was very simple code.

The seeding is small, fast and extremely difficult to duplicate yet there are many bigger, slower techniques that could be use, get the result from GetTickCount and BSWAP it but that involves an API call. There are many other methods but I doubt they are that small or as fast as the RDTSC BSWAP combination.

Now as far as "flogging to death", how many unintelligible acronyms have I heard out of you over the last few years ?

As far as the choice of mnemonics, your whining seems to have missed the rest of the mnemonics.

  MOV
  ADD
  CMP
  JNE

I can assure you I HAVE flogged all of these but not to death, just mercilessly for years and there are many others.

Now the only reason I can think of is that both you and Stuart are suffering lockdown boredom, can't you find a good movie to watch, roll on down to your local pub and do injustice to a few pints, go for a walk in the park or something productive ?

[David Roberts]

Thanks, Dale. I have had this jpg on my desktop since I joined FreeBASIC.



Stuart:

It is funny how the brain works. We don't buy into a statement. We don't buy into another statement. Another statement comes along and we buy into it.

The BSWAP is superfluous in this use case

You have persuaded me.

[Steve Hutchesson]

Here is a variation that should keep you happy as long as you don't look at the macro.

' ¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤

  MACRO Dont_Call_This_Reseeding
    MACROTEMP nvar
    LOCAL nvar as DWORD
    ! rdtsc
    ! bswap eax
    ! mov nvar, eax
    randomize nvar
  END MACRO

FUNCTION randstr(ByVal ccnt as DWORD) as STRING

    #REGISTER NONE

    LOCAL cnt as DWORD

    Dont_Call_This_Reseeding

    src$ = ""

    cnt = 0
    ! mov esi, ccnt

  lbl:
    src$ = src$ + chr$(rnd(0,255))
    ! add cnt, 1
    ! cmp cnt, esi
    ! jne lbl

    FUNCTION = src$

End FUNCTION

' ¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤

[Stuart McLachlan]

OK, I finally looked at the algorithm link.

WTF?
Using RND(0 to 255) to generate a Windows class name (NULL-TERMINATED) string?

1 in 256 times, you generate a NULL string.
I'll leave it up to others to work out how many times you end up with a 1,2 or 3 character string

Added: Ok, I decided to do it myself;
'

#COMPILE EXE
FUNCTION PBMAIN() AS LONG
    LOCAL x,y AS LONG
    LOCAL q AS QUAD
    LOCAL s AS STRING
    LOCAL sWinName AS STRINGZ*65
    LOCAL hTxt AS DWORD
    DIM aLen(0 TO 64) AS LONG
    TIX q
    RANDOMIZE LO(DWORD,q)
    FOR x = 1 TO 10000
        s=""
        FOR y = 1 TO 64
            s+=CHR$(RND(0,255))
        NEXT
        sWinName = s
        INCR aLen(LEN(sWinName))
     NEXT
    s="Len  Count" & $LF
    FOR x = 0 TO 3
        s += STR$(x) & $TAB & STR$(aLen(x)) & $LF
    NEXT
    ? s

END FUNCTION
'

[Steve Hutchesson]

Stuart,

> WIth a register that is rotating through the 2^32 values in the DWORD about every 2 seconds and a single request for a seed for the RNG, what is the "sequential" aspect you are so concerned about?

You have misunderstood BSWAP. It does not rotate the register, it reverses the byte order.

00112233 becomes 33221100

[Steve Hutchesson]

WTF?
Using CHR$(0 to 255) to generate a Windows class name (NULL-TERMINATED) string?

WTF,
The most you can lose is one trailing zero and remember, its being written to a basic dynamic string that can contain ascii zeros.

[David Roberts]

You appear to have missed what the posted algo does

As John Wayne would say: "The hell I did!"

how many unintelligible acronyms have I heard out of you over the last few years ?

Unintelligible acronyms? None. If ever I use an acronym, like NIST for example, I would write National Institute of Standards and Technology (NIST) and then use NIST thereafter. I think you meant something else - not acronyms. Perhaps you meant metaphor. I am from Yorkshire, England, and we are very metaphorical. Have a conversation with a Yorky and it is like talking to a poet. Having said that, you'd probably end up scratching your head until the cows come home.

your whining seems to have missed the rest of the mnemonics

Whining? Your perception is shot to pieces.

Now the only reason I can think of is that both you and Stuart are suffering lockdown boredom ...

If that is the only reason I suggest that you go for a walk in the park and give it a bit more thought.

I looked at your post #16 and it did not do a lot for me.

as long as you don't look at the macro.

I don't have a major issue with your idea - never did have. It is just another tool in the toolbox. As I keep saying I use a CPRNG to seed a PRNG. I use an API - so what? If two pieces of code are below the noise floor it matters not which is the fastest.