[adrian_b]

C had 3 major sources, B (derived from BCPL, which had been derived from CPL, which had been derived from ALGOL 60), IBM PL/I and ALGOL 68.

Structs come from PL/I, not from ALGOL 68, together with the postfix operators "." and "->". The term "pointer" also comes from PL/I, the corresponding term in ALGOL 68 was "reference". The prefix operator "*" is a mistake peculiar to C, acknowledged later by the C language designers, it should have been a postfix operator, like in Euler and Pascal.

Examples of things that come from ALGOL 68 are unions (unfortunately C unions lack most useful features of the ALGOL 68 unions. which are implicitly tagged unions) and the combined operation-assignment operators, e.g. "+=" or "*=".

The Bourne shell scripting language, inherited by ksh, bash, zsh etc., also has many features taken from ALGOL 68.

The explicit "malloc" and "free" also come from PL/I. ALGOL 68 is normally implemented with a garbage collector.

[Taniwha]

C originally had =+ and =- (upto and including Unix V6) - they were ambiguous (a=-b means a= -b? or a = a-b?) and replaced by +=/-=

The original structs were pretty bad too - field names had their own address space and could sort of be used with any pointer which sort of allowed you to make tacky unions) we didn't get a real type system until the late 80s

[adrian_b]

ALGOL 68 had "=" for equality and ":=" for assignment, like ALGOL 60.

Therefore the operation with assignment operators were like "+:=".

The initial syntax of C was indeed weird and it was caused by the way how their original parser in their first C compiler happened to be written and rewritten, the later form of the assignment operators was closer to their source from ALGOL 68.

[ErikCorry]

Yeah if you ever wondered why the fields in a lot of Posix APIs have names with prefixes like tm_sec and tm_usec it's because of this misfeature of early C.

[themafia]

> it should have been a postfix operator, like in Euler and Pascal.

I never liked Pascal style Pointer^. As the postfix starts to get visually cumbersome with more than one layer of Indirection^^. Especially when combined with other postfix Operators^^.AndMethods. Or even just Operator^ := Assignment.

I also think it's the natural inverse of the "address-of" prefix operator. So we have "take the address of this value" and "look through the address to retreive the value."

[adrian_b]

The "natural inverse" relationship between "address-of" and indirect addressing is only partial.

You can apply the "*" operator as many times you want, but applying "address-of" twice is meaningless.

Moreover, in complex expressions it is common to mix the indirection operator with array indexing and with structure member selection, and all these 3 postfix operators can appear an unlimited number of times in an expression.

Writing such addressing expressions in C is extremely cumbersome, because they require a great number of parentheses levels and it is still difficult to see which is the order in which they are applied.

With a postfix indirection operator no parentheses are needed and all addressing operators are executed in the order in which they are written.

So it is beyond reasonable doubt that a prefix "*" is a mistake.

The only reason why they have chosen "*" as prefix in C, which they later regretted, was because it seemed easier to define the expressions "*++p" and "*p++" to have the desired order of evaluation.

There is no other use case where a prefix "*" simplifies anything and for the postfix and prefix increment and decrement it would have been possible to find other ways to avoid parentheses and even if they were used with parentheses that would still have been simpler than when you have to mix "*" with array indexing and with structure member selection. Moreover, the use of "++" and "--" with pointers was only a workaround for a dumb compiler, which could not determine by itself whether it should access an array using indices or pointers. Normally there should be no need to expose such an implementation detail in a high-level language, the compiler should choose the addressing modes that are optimal for the target CPU, not the programmer. On some CPUs, including the Intel/AMD CPUs, accessing arrays by incrementing pointers, like in the old C programs, is usually worse than accessing the arrays through indices (because on such CPUs the loop counter can be reused as an index register, regardless of the order in which the array is accessed, including for accessing multiple arrays, avoiding the use of extra registers and reducing the number of executed instructions).

With a postfix "*", the operator "->" would have been superfluous. It has been added to C only to avoid some of the most frequent cases when a prefix "*" leads to ugly syntax.

[themafia]

> You can apply the "*" operator as many times you want, but applying "address-of" twice is meaningless.

This is due to the nature of lvalue and rvalue expressions. You can only get an object where * is meaningful twice if you've applied & meaningfully twice before.

    int a = 42;
    int *b = &a;
    int **c = &b;

I've applied & twice. I merely had to negotiate with the language instead of the parser to do so.

> and all these 3 postfix operators can appear an unlimited number of times in an expression.

In those cases the operator is immediately followed by a non-operator token. I cannot meaningfully write a[][1], or b..field.

> The only reason why they have chosen "*" as prefix in C, which they later regretted, was because it seemed easier to define the expressions "++p" and "p++" to have the desired order of evaluation.

It not only seems easier it is easier. What you sacrifice is complications is defining function pointers. One is far more common than the other. I think they got it right.

> With a postfix "*", the operator "->" would have been superfluous.

Precisely the reason I dislike the Pascal**.Style. Go offers a better mechanism anyways. Just use "." and let the language work out what that means based on types.

I'm offering a subjective point of view. I don't like the way that looks or reads or mentally parses. I'm much happier to occasionally struggle with function pointers.

[LeFantome]

I do not think that is what they meant.

**c is valid but &&b makes no sense.

[comex]

Some languages do define &&b, like Rust, where its effect is similar to the parent post's C example: it creates a temporary stack allocation initialized with &b, and then takes the address of that.

You could argue this is inconsistent or confusing. It is certainly useful though.

Incidentally, C99 lets you do something similar with compound literal syntax; this is a valid expression:

    &(int *){&b}

[inkyoto]

> The only reason why they have chosen "" as prefix in C, which they later regretted, was because it seemed easier to define the expressions "++p" and "*p++" to have the desired order of evaluation.

There has been no shortage of speculation, much of it needlessly elaborate. The reality, however, appears far simpler – the prefix pointer notation had already been present in B and its predecessor, BCPL[0]. It was not invented anew, merely borrowed – or, more accurately, inherited.

The common lore often attributes this syntactic feature to the influence of the PDP-11 ISA. That claim, whilst not entirely baseless, is at best a partial truth. The PDP-11 did support pre-increment and post-increment indirect address manipulation – but notably lacked their symmetrical complements: pre-increment and post-decrement addressing modes[1]. In other words, it exhibited asymmetry – a gap that undermines the argument for direct PDP-11 ISA inheritance, i.e.

  MOV (Rn)+, Rm

  MOV @(Rn)+, Rm

  MOV -(Rn), Rm

  MOV @-(Rn), Rm

existed but not

  MOV +(Rn), Rm

  MOV @+(Rn), Rm

  MOV (Rn)-, Rm

  MOV @(Rn)-, Rm

[0] https://www.thinkage.ca/gcos/expl/b/manu/manu.html#Section6_...

[1] PDP-11 ISA allocates 3 bits for the addressing mode (register / Rn, indirect register (Rn), auto post-increment indirect / (Rn)+ , auto post-increment deferred / @(Rn)+, auto pre-decrement indirect / -(Rn), auto pre-increment deferred / @-(Rn), index / idx(Rn) and index deferred / @idx(Rn) ), and whether it was actually «let's choose these eight modes» or «we also wanted pre-increment and post-decrement but ran out of bits» is a matter of historical debate.

[adrian_b]

The prefix "*" and the increment/decrement operators have been indeed introduced in the B language (in 1969, before the launch of PDP-11 in 1970, but earlier computers had some autoincrement/autodecrement facilities, though not as complete as in the B language), where "*" has been made prefix for the reason that I have already explained.

The prefix "*" WAS NOT inherited from BCPL, it was purely a B invention due to Ken Thompson.

In BCPL, "*" was actually a postfix operator that was used for array indexing. It was not the operator for indirection.

In CPL, the predecessor of BCPL, there was no indirection operator, because indirection through a pointer was implicit, based on the type of the variable. Instead of an indirection operator, there were different kinds of assignment operators, to enable the assignment of a value to the pointer, instead of assigning to the variable pointed by the pointer, which was the default meaning.

BCPL has made many changes in the syntax of CPL, whose main reason was the necessity of adapting the language to the impoverished character set available on American computers, which lacked many of the characters that had been available in Europe before IBM and a few other US vendors have succeeded to replace the local vendors, also imposing thus the EBCDIC and later the ASCII character sets.

Several of the changes done between BCPL and B had the same kind of reason, i.e. they were needed to transition the language from an older character set to the then new ASCII character set. For instance the use of braces as block delimiters was prompted by their addition into ASCII, as they were not available in the previous character set.

The link that you have provided to a manual of the B language is not useful for historical discussions, as the manual is for a modernized version of B, which contains some features back-ported from C.

There is a manual of the B language dated 1972-01-07, which predates the C language, and which can be found on the Web. Even that version might have already included some changes from the original B language of 1969.

[inkyoto]

* was the usual infix multiplication operator in BCPL, and it was not used for pointer arithmetic.

The BCPL manual[0] explains the «monadic !» operator (section 2.11.3) as:

  2.11.3 MONADIC !

  The value or a monadic ! expression is the value of the storage cell whose address is the operand of the !. Thus @!E = !@E = E, (providing E is an expression of the class described in 2.11.2).

  Examples.

  !X := Y Stores the value of Y into the storage cell whose address is the value of X.

  P := !P Stores the value of the cell whose address is the value of P, as the new value of P.

The array indexing used the «V ! idx» syntax (section 2.13, «Vector application»).

So, the ! was a prefix operator for pointers, and it was an infix operator for array indexing.

In Richard's account of BCPL's evolution, he noted that on early hardware the exlamation mark was not easily available, and, therefore, he used a composite *( (i.e. a diagraph):

  «The star in *( was chosen because it was available … and it seemed appropriate for subscription since it was used as the indirection operator in the FAP assembly language on CTSS. Later, when the exclamation mark became available, *( was replaced by !( and exclamation mark became both a dyadic and monadic indirection operator».

So, in all likelihood, !X := Y became *(X := Y, eventually becoming *X = Y (in B and C) whilst retaining the exact and original semantics of the !.

[0] https://rabbit.eng.miami.edu/info/bcpl_reference_manual.pdf

[adrian_b]

The BCPL manual linked by you is not useful, as it describes a recent version of the language, which is irrelevant for the evolution of the B and C languages. A manual of BCPL from July 1967, predating B, can be found on the Web.

The use of the character "!" in BCPL is much later than the development of the B language from BCPL, in 1969.

The asterisk had 3 uses in BCPL, as the multiplication operator, as a marker for the opening bracket in array indexing, to compensate for the lack of different kinds of brackets for function evaluation and for array indexing, and as the escape character in character strings. For the last use the asterisk has been replaced by the backslash in C.

There was indeed a prefix indirection operator in BCPL, but it did not use any special character, because the available character set did not have any unused characters.

The BCPL parser was separate from the lexer, and it was possible for the end users to modify the lexer, in order to assign any locally available characters to the syntactic tokens.

So if a user had appropriate characters, they could have been assigned to indirection and address-of, but otherwise they were just written RV and LV, for right-hand-side value and left-hand-side value.

It is not known whether Ken Thompson had modified the BCPL lexer for his PDP computer, to use some special characters for operators like RV and LV.

In any case, he could not have used asterisk for indirection, because that would have conflicted with its other uses.

The use of asterisk for indirection in B became possible only after Ken Thompson has made many other changes and simplifications in comparison with BCPL, removing any parsing conflicts.

You are right that BCPL already had prefix operators for indirection and address-of, which was different from how this had been handled in CPL, but Martin Richards did not seem to have any reason for this choice and in BCPL this was a less obvious mistake, because it did not have structures.

On the other hand, Ken Thompson did want to have "*" as prefix, after introducing his increment and decrement operators, in order to need no parentheses for pre- and post-incrementation or decrementation of pointers, in the context where postfix operators were defined as having higher precedence than prefix.

Also in his case this was not yet an obvious mistake, because he had no structures and the programs written in B at that time did not use any complex data structures that would need correspondingly complex addressing expressions.

Only years later it became apparent that this was a bad choice, while the earlier choice of N. Wirth in Euler (January 1966; the first high-level language that handled pointers explicitly, with indirection and address-of operators) had been the right one. The high-level languages that had "references" before 1966 (the term "pointer" has been introduced in IBM PL/I, in July 1966), e.g. CPL and FORTRAN IV, handled them only implicitly.

Decades later, complex data structures became common while the manual optimization of incrementing/decrementing explicitly pointers for addressing arrays became a way of writing inefficient programs, which prevent the compiler from optimizing correctly the array accessing for the target CPU.

So the choice of Ken Thompson can be justified in its context from 1969, but in hindsight it has definitely been a very bad choice.

[zozbot234]

A postfix "*" would be completely redundant since you can just use p[0] . Instead of *p++ you'd have (p++)[0] - still quite workable.

[fastaguy88]

You're kidding, right? (p++)[0] returns the contents of (p) before the ++. Its hard to imagine a more confusing juxtaposition.

[psychoslave]

A dash instead of a dot would be so much more congruent with the way Latin script generally render compounded terms. And a reference/pointer (or even pin for short) is really nothing that much different compared to any other function/operator/method.

some·object-pin-pin-pin-transform is not harder to parse nor to interpret as human than (***some_object)->transform().

[inkyoto]

C's «static» and «auto» also come from PL/I. Even though «auto» has never been used in C, it has found its place in C++.

C also had a reserved keyword, «entry», which had never been used before eventually being relinquished from its keyword status when the standardisation of C began.

[pjmlp]

C23 also has reused auto as C++, although type inference is more limited.