V0.1 Similarities and differences with Go

Agora is syntactically very similar to Go. But its goal is obviously not to be a clone, so there are also important differences. Most of those differences support the goal of making a looser, dynamic and simple interpreted language that complements Go. This article documents these similarities and differences.

Similarities

• The first and most immediately obvious similarity is the syntax. Like Go, Agora uses curly braces to delimit code blocks, and is semicolon-free (semicolons are optional and are automatically added at the scanner stage of the compiler - like Go).

• The agora keywords are all mostly the same as the Go keywords - although agora has less. They are if, else, for, func, return, debug, break and continue. Of these, only debug is agora-specific.

• Many operators are also the same. They are +, -, *, /, %, ==, !=, >, <, >=, <=, !, the logical operators && and ||, the increment and decrement ++ and --, and the assignment operators =, :=, +=, -=, *=, /=, %=. Operator precedence is also the same as Go.

• Field access can be done via the dot . operator, like Go.

• Function calls use parentheses around the parameter list (even if there is no parameter), while if and for statements don't require parentheses.

• The nil literal represents a null value, and boolean literals are true and false as expected.

• len is a built-in that returns the length of a string, or the number of items in an object (roughly similar to a map).

• import, panic and recover are present and do mostly what is expected, but with subtle differences, see the next section.

• Comments use the same notation as Go, namely // for one-liners, and /* */ for multi-line.

• Cyclic dependencies are prohibited, they are detected and raise an error (at runtime).

Differences

• The most obvious and important difference is probably the lack of types. Variables, arguments and functions have no types. Values have a type (a value is either a string, a number, a bool, an object, a func or nil).

• Code is organized in single-file modules. Statements can (and must, essentially!) appear at the top-level of the module. This is implicitly the top-level function of the module, the code that gets executed when the module is imported.

• There is no uppercase rule for what is or isn't exported by a module. The only thing a module exports is the return value of the top-level function. Anything else is private inside the module. To return multiple functions (say, like the fmt module), an object is returned. However, the uppercase-is-public is still used as a convention (for instance, all stdlib's functions start with an uppercase).

• There is no "short statement" allowed in an if statement before the condition.

• There is a ternary condition ? iftrue : iffalse operator.

• A function may receive an arbitrary number of arguments, maybe exceeding the number of expected (formal) arguments declared in its signature. All actual arguments passed to a function are always available using the reserved args identifier, which is an array-like object, with keys ranging from 0 to the number of arguments received minus one. Also, since the top-level function cannot declare expected arguments, this is the only way to retrieve arguments passed to the module.

• Unlike Go, import is a built-in function, not a keyword that must appear at the top of the package. So it can be called wherever makes most sense, since this can be a costly operation (loading from a file, compiling, executing). As mentioned previously, it returns the value returned by the module and must be stored in a variable (there is no implicit "variable" derived from the import path).

• There are no multiple return values, functions can only return a single value.

• There are truthy and falsy values other than the boolean true and false. Namely, the 0 number, the empty string, nil and false are all false. An object can also be false if it provides a meta-method __bool.

• The panic built-in function takes a value and raises an error with it. However, it doesn't raise if the value is falsy. This is symmetric with the behaviour of recover.

• The recover built-in function takes a function as parameter, and executes it in protected mode. By default, a runtime error is a panic, and stops all agora code execution to return the error to the Go host program (the second value returned from module.Run()). To catch errors in agora code, the recover function must be used, this is what is called the protected mode (similar to Lua error handling). recover runs the provided function, and if an error occurs, it catches it and returns it. Otherwise it returns nil.

• Field access can also be done using an array-like syntax, object["key"] = value. Any type except nil can be used as the key, and when assigning to a field using the . operator, the key is implicitly a string (denoted by the identifier of the field), so object.key is equivalent to object["key"].

• There are no bitwise operators.

• There is no multi-thread (goroutine) or channel support. Agora code must be single-threaded, although different execution contexts can be run in parallel.

• There are no slices or maps, the only compound data structure is the object. It can represent a slice and a map, and eventually will be optimized for when used as a slice/array.

Next: Language reference