Noteworthy Differences from other Languages

Noteworthy differences from MATLAB

Although MATLAB users may find Julia's syntax familiar, Julia is not a MATLAB clone. There are major syntactic and functional differences. The following are some noteworthy differences that may trip up Julia users accustomed to MATLAB:

  • Julia arrays are indexed with square brackets, A[i,j].
  • Julia arrays are not copied when assigned to another variable. After A = B, changing elements of B will modify A as well.
  • Julia values are not copied when passed to a function. If a function modifies an array, the changes will be visible in the caller.
  • Julia does not automatically grow arrays in an assignment statement. Whereas in MATLAB a(4) = 3.2 can create the array a = [0 0 0 3.2] and a(5) = 7 can grow it into a = [0 0 0 3.2 7], the corresponding Julia statement a[5] = 7 throws an error if the length of a is less than 5 or if this statement is the first use of the identifier a. Julia has push! and append!, which grow Vectors much more efficiently than MATLAB's a(end+1) = val.
  • The imaginary unit sqrt(-1) is represented in Julia as im, not i or j as in MATLAB.
  • In Julia, literal numbers without a decimal point (such as 42) create integers instead of floating point numbers. As a result, some operations can throw a domain error if they expect a float; for example, julia> a = -1; 2^a throws a domain error, as the result is not an integer (see the FAQ entry on domain errors for details).
  • In Julia, multiple values are returned and assigned as tuples, e.g. (a, b) = (1, 2) or a, b = 1, 2. MATLAB's nargout, which is often used in MATLAB to do optional work based on the number of returned values, does not exist in Julia. Instead, users can use optional and keyword arguments to achieve similar capabilities.
  • Julia has true one-dimensional arrays. Column vectors are of size N, not Nx1. For example, rand(N) makes a 1-dimensional array.
  • In Julia, [x,y,z] will always construct a 3-element array containing x, y and z.
    • To concatenate in the first ("vertical") dimension use either vcat(x,y,z) or separate with semicolons ([x; y; z]).
    • To concatenate in the second ("horizontal") dimension use either hcat(x,y,z) or separate with spaces ([x y z]).
    • To construct block matrices (concatenating in the first two dimensions), use either hvcat or combine spaces and semicolons ([a b; c d]).
  • In Julia, a:b and a:b:c construct AbstractRange objects. To construct a full vector like in MATLAB, use collect(a:b). Generally, there is no need to call collect though. An AbstractRange object will act like a normal array in most cases but is more efficient because it lazily computes its values. This pattern of creating specialized objects instead of full arrays is used frequently, and is also seen in functions such as range, or with iterators such as enumerate, and zip. The special objects can mostly be used as if they were normal arrays.
  • Functions in Julia return values from their last expression or the return keyword instead of listing the names of variables to return in the function definition (see The return Keyword for details).
  • A Julia script may contain any number of functions, and all definitions will be externally visible when the file is loaded. Function definitions can be loaded from files outside the current working directory.
  • In Julia, reductions such as sum, prod, and max are performed over every element of an array when called with a single argument, as in sum(A), even if A has more than one dimension.
  • In Julia, parentheses must be used to call a function with zero arguments, like in rand().
  • Julia discourages the use of semicolons to end statements. The results of statements are not automatically printed (except at the interactive prompt), and lines of code do not need to end with semicolons. println or @printf can be used to print specific output.
  • In Julia, if A and B are arrays, logical comparison operations like A == B do not return an array of booleans. Instead, use A .== B, and similarly for the other boolean operators like <, >.
  • In Julia, the operators &, |, and (xor) perform the bitwise operations equivalent to and, or, and xor respectively in MATLAB, and have precedence similar to Python's bitwise operators (unlike C). They can operate on scalars or element-wise across arrays and can be used to combine logical arrays, but note the difference in order of operations: parentheses may be required (e.g., to select elements of A equal to 1 or 2 use (A .== 1) .| (A .== 2)).
  • In Julia, the elements of a collection can be passed as arguments to a function using the splat operator ..., as in xs=[1,2]; f(xs...).
  • Julia's svd returns singular values as a vector instead of as a dense diagonal matrix.
  • In Julia, ... is not used to continue lines of code. Instead, incomplete expressions automatically continue onto the next line.
  • In both Julia and MATLAB, the variable ans is set to the value of the last expression issued in an interactive session. In Julia, unlike MATLAB, ans is not set when Julia code is run in non-interactive mode.
  • Julia's structs do not support dynamically adding fields at runtime, unlike MATLAB's classes. Instead, use a Dict. Dict in Julia isn't ordered.
  • In Julia each module has its own global scope/namespace, whereas in MATLAB there is just one global scope.
  • In MATLAB, an idiomatic way to remove unwanted values is to use logical indexing, like in the expression x(x>3) or in the statement x(x>3) = [] to modify x in-place. In contrast, Julia provides the higher order functions filter and filter!, allowing users to write filter(z->z>3, x) and filter!(z->z>3, x) as alternatives to the corresponding transliterations x[x.>3] and x = x[x.>3]. Using filter! reduces the use of temporary arrays.
  • The analogue of extracting (or "dereferencing") all elements of a cell array, e.g. in vertcat(A{:}) in MATLAB, is written using the splat operator in Julia, e.g. as vcat(A...).
  • In Julia, the adjoint function performs conjugate transposition; in MATLAB, adjoint provides the "adjugate" or classical adjoint, which is the transpose of the matrix of cofactors.
  • In Julia, a^b^c is evaluated a^(b^c) while in MATLAB it's (a^b)^c.

Noteworthy differences from R

One of Julia's goals is to provide an effective language for data analysis and statistical programming. For users coming to Julia from R, these are some noteworthy differences:

  • Julia's single quotes enclose characters, not strings.

  • Julia can create substrings by indexing into strings. In R, strings must be converted into character vectors before creating substrings.

  • In Julia, like Python but unlike R, strings can be created with triple quotes """ ... """. This syntax is convenient for constructing strings that contain line breaks.

  • In Julia, varargs are specified using the splat operator ..., which always follows the name of a specific variable, unlike R, for which ... can occur in isolation.

  • In Julia, modulus is mod(a, b), not a %% b. % in Julia is the remainder operator.

  • Julia constructs vectors using brackets. Julia's [1, 2, 3] is the equivalent of R's c(1, 2, 3).

  • In Julia, not all data structures support logical indexing. Furthermore, logical indexing in Julia is supported only with vectors of length equal to the object being indexed. For example:

    • In R, c(1, 2, 3, 4)[c(TRUE, FALSE)] is equivalent to c(1, 3).
    • In R, c(1, 2, 3, 4)[c(TRUE, FALSE, TRUE, FALSE)] is equivalent to c(1, 3).
    • In Julia, [1, 2, 3, 4][[true, false]] throws a BoundsError.
    • In Julia, [1, 2, 3, 4][[true, false, true, false]] produces [1, 3].
  • Like many languages, Julia does not always allow operations on vectors of different lengths, unlike R where the vectors only need to share a common index range. For example, c(1, 2, 3, 4) + c(1, 2) is valid R but the equivalent [1, 2, 3, 4] + [1, 2] will throw an error in Julia.

  • Julia allows an optional trailing comma when that comma does not change the meaning of code. This can cause confusion among R users when indexing into arrays. For example, x[1,] in R would return the first row of a matrix; in Julia, however, the comma is ignored, so x[1,] == x[1], and will return the first element. To extract a row, be sure to use :, as in x[1,:].

  • Julia's map takes the function first, then its arguments, unlike lapply(<structure>, function, ...) in R. Similarly Julia's equivalent of apply(X, MARGIN, FUN, ...) in R is mapslices where the function is the first argument.

  • Multivariate apply in R, e.g. mapply(choose, 11:13, 1:3), can be written as broadcast(binomial, 11:13, 1:3) in Julia. Equivalently Julia offers a shorter dot syntax for vectorizing functions binomial.(11:13, 1:3).

  • Julia uses end to denote the end of conditional blocks, like if, loop blocks, like while/ for, and functions. In lieu of the one-line if ( cond ) statement, Julia allows statements of the form if cond; statement; end, cond && statement and !cond || statement. Assignment statements in the latter two syntaxes must be explicitly wrapped in parentheses, e.g. cond && (x = value).

  • In Julia, <-, <<- and -> are not assignment operators.

  • Julia's -> creates an anonymous function.

  • Julia's * operator can perform matrix multiplication, unlike in R. If A and B are matrices, then A * B denotes a matrix multiplication in Julia, equivalent to R's A %*% B. In R, this same notation would perform an element-wise (Hadamard) product. To get the element-wise multiplication operation, you need to write A .* B in Julia.

  • Julia performs matrix transposition using the transpose function and conjugated transposition using the ' operator or the adjoint function. Julia's transpose(A) is therefore equivalent to R's t(A). Additionally a non-recursive transpose in Julia is provided by the permutedims function.

  • Julia does not require parentheses when writing if statements or for/while loops: use for i in [1, 2, 3] instead of for (i in c(1, 2, 3)) and if i == 1 instead of if (i == 1).

  • Julia does not treat the numbers 0 and 1 as Booleans. You cannot write if (1) in Julia, because if statements accept only booleans. Instead, you can write if true, if Bool(1), or if 1==1.

  • Julia does not provide nrow and ncol. Instead, use size(M, 1) for nrow(M) and size(M, 2) for ncol(M).

  • Julia is careful to distinguish scalars, vectors and matrices. In R, 1 and c(1) are the same. In Julia, they cannot be used interchangeably.

  • Julia's diag and diagm are not like R's.

  • Julia cannot assign to the results of function calls on the left hand side of an assignment operation: you cannot write diag(M) = fill(1, n).

  • Julia discourages populating the main namespace with functions. Most statistical functionality for Julia is found in packages under the JuliaStats organization. For example:

  • Julia provides tuples and real hash tables, but not R-style lists. When returning multiple items, you should typically use a tuple or a named tuple: instead of list(a = 1, b = 2), use (1, 2) or (a=1, b=2).

  • Julia encourages users to write their own types, which are easier to use than S3 or S4 objects in R. Julia's multiple dispatch system means that table(x::TypeA) and table(x::TypeB) act like R's table.TypeA(x) and table.TypeB(x).

  • In Julia, values are not copied when assigned or passed to a function. If a function modifies an array, the changes will be visible in the caller. This is very different from R and allows new functions to operate on large data structures much more efficiently.

  • In Julia, vectors and matrices are concatenated using hcat, vcat and hvcat, not c, rbind and cbind like in R.

  • In Julia, a range like a:b is not shorthand for a vector like in R, but is a specialized AbstractRange object that is used for iteration. To convert a range into a vector, use collect(a:b).

  • The : operator has a different precedence in R and Julia. In particular, in Julia arithmetic operators have higher precedence than the : operator, whereas the reverse is true in R. For example, 1:n-1 in Julia is equivalent to 1:(n-1) in R.

  • Julia's max and min are the equivalent of pmax and pmin respectively in R, but both arguments need to have the same dimensions. While maximum and minimum replace max and min in R, there are important differences.

  • Julia's sum, prod, maximum, and minimum are different from their counterparts in R. They all accept an optional keyword argument dims, which indicates the dimensions, over which the operation is carried out. For instance, let A = [1 2; 3 4] in Julia and B <- rbind(c(1,2),c(3,4)) be the same matrix in R. Then sum(A) gives the same result as sum(B), but sum(A, dims=1) is a row vector containing the sum over each column and sum(A, dims=2) is a column vector containing the sum over each row. This contrasts to the behavior of R, where separate colSums(B) and rowSums(B) functions provide these functionalities. If the dims keyword argument is a vector, then it specifies all the dimensions over which the sum is performed, while retaining the dimensions of the summed array, e.g. sum(A, dims=(1,2)) == hcat(10). It should be noted that there is no error checking regarding the second argument.

  • Julia has several functions that can mutate their arguments. For example, it has both sort and sort!.

  • In R, performance requires vectorization. In Julia, almost the opposite is true: the best performing code is often achieved by using devectorized loops.

  • Julia is eagerly evaluated and does not support R-style lazy evaluation. For most users, this means that there are very few unquoted expressions or column names.

  • Julia does not support the NULL type. The closest equivalent is nothing, but it behaves like a scalar value rather than like a list. Use x === nothing instead of is.null(x).

  • In Julia, missing values are represented by the missing object rather than by NA. Use ismissing(x) (or ismissing.(x) for element-wise operation on vectors) instead of is.na(x). The skipmissing function is generally used instead of na.rm=TRUE (though in some particular cases functions take a skipmissing argument).

  • Julia lacks the equivalent of R's assign or get.

  • In Julia, return does not require parentheses.

  • In R, an idiomatic way to remove unwanted values is to use logical indexing, like in the expression x[x>3] or in the statement x = x[x>3] to modify x in-place. In contrast, Julia provides the higher order functions filter and filter!, allowing users to write filter(z->z>3, x) and filter!(z->z>3, x) as alternatives to the corresponding transliterations x[x.>3] and x = x[x.>3]. Using filter! reduces the use of temporary arrays.

Noteworthy differences from Python

  • Julia's for, if, while, etc. blocks are terminated by the end keyword. Indentation level is not significant as it is in Python. Unlike Python, Julia has no pass keyword.
  • Strings are denoted by double quotation marks ("text") in Julia (with three double quotation marks for multi-line strings), whereas in Python they can be denoted either by single ('text') or double quotation marks ("text"). Single quotation marks are used for characters in Julia ('c').
  • String concatenation is done with * in Julia, not + like in Python. Analogously, string repetition is done with ^, not *. Implicit string concatenation of string literals like in Python (e.g. 'ab' 'cd' == 'abcd') is not done in Julia.
  • Python Lists—flexible but slow—correspond to the Julia Vector{Any} type or more generally Vector{T} where T is some non-concrete element type. "Fast" arrays like NumPy arrays that store elements in-place (i.e., dtype is np.float64, [('f1', np.uint64), ('f2', np.int32)], etc.) can be represented by Array{T} where T is a concrete, immutable element type. This includes built-in types like Float64, Int32, Int64 but also more complex types like Tuple{UInt64,Float64} and many user-defined types as well.
  • In Julia, indexing of arrays, strings, etc. is 1-based not 0-based.
  • Julia's slice indexing includes the last element, unlike in Python. a[2:3] in Julia is a[1:3] in Python.
  • Unlike Python, Julia allows AbstractArrays with arbitrary indexes. Python's special interpretation of negative indexing, a[-1] and a[-2], should be written a[end] and a[end-1] in Julia.
  • Julia requires end for indexing until the last element. x[1:] in Python is equivalent to x[2:end] in Julia.
  • Julia's range indexing has the format of x[start:step:stop], whereas Python's format is x[start:(stop+1):step]. Hence, x[0:10:2] in Python is equivalent to x[1:2:10] in Julia. Similarly, x[::-1] in Python, which refers to the reversed array, is equivalent to x[end:-1:1] in Julia.
  • In Julia, ranges can be constructed independently as start:step:stop, the same syntax it uses in array-indexing. The range function is also supported.
  • In Julia, indexing a matrix with arrays like X[[1,2], [1,3]] refers to a sub-matrix that contains the intersections of the first and second rows with the first and third columns. In Python, X[[1,2], [1,3]] refers to a vector that contains the values of cell [1,1] and [2,3] in the matrix. X[[1,2], [1,3]] in Julia is equivalent with X[np.ix_([0,1],[0,2])] in Python. X[[0,1], [0,2]] in Python is equivalent with X[[CartesianIndex(1,1), CartesianIndex(2,3)]] in Julia.
  • Julia has no line continuation syntax: if, at the end of a line, the input so far is a complete expression, it is considered done; otherwise the input continues. One way to force an expression to continue is to wrap it in parentheses.
  • Julia arrays are column-major (Fortran-ordered) whereas NumPy arrays are row-major (C-ordered) by default. To get optimal performance when looping over arrays, the order of the loops should be reversed in Julia relative to NumPy (see relevant section of Performance Tips).
  • Julia's updating operators (e.g. +=, -=, ...) are not in-place whereas NumPy's are. This means A = [1, 1]; B = A; B += [3, 3] doesn't change values in A, it rather rebinds the name B to the result of the right-hand side B = B + 3, which is a new array. For in-place operation, use B .+= 3 (see also dot operators), explicit loops, or InplaceOps.jl.
  • Julia evaluates default values of function arguments every time the method is invoked, unlike in Python where the default values are evaluated only once when the function is defined. For example, the function f(x=rand()) = x returns a new random number every time it is invoked without argument. On the other hand, the function g(x=[1,2]) = push!(x,3) returns [1,2,3] every time it is called as g().
  • In Julia, keyword arguments must be passed using keywords, unlike Python in which it is usually possible to pass them positionally. Attempting to pass a keyword argument positionally alters the method signature leading to a MethodError or calling of the wrong method.
  • In Julia % is the remainder operator, whereas in Python it is the modulus.
  • In Julia, the commonly used Int type corresponds to the machine integer type (Int32 or Int64), unlike in Python, where int is an arbitrary length integer. This means in Julia the Int type will overflow, such that 2^64 == 0. If you need larger values use another appropriate type, such as Int128, BigInt or a floating point type like Float64.
  • The imaginary unit sqrt(-1) is represented in Julia as im, not j as in Python.
  • In Julia, the exponentiation operator is ^, not ** as in Python.
  • Julia uses nothing of type Nothing to represent a null value, whereas Python uses None of type NoneType.
  • In Julia, the standard operators over a matrix type are matrix operations, whereas, in Python, the standard operators are element-wise operations. When both A and B are matrices, A * B in Julia performs matrix multiplication, not element-wise multiplication as in Python. A * B in Julia is equivalent with A @ B in Python, whereas A * B in Python is equivalent with A .* B in Julia.
  • The adjoint operator ' in Julia returns an adjoint of a vector (a lazy representation of row vector), whereas the transpose operator .T over a vector in Python returns the original vector (non-op).
  • In Julia, a function may contain multiple concrete implementations (called methods), which are selected via multiple dispatch based on the types of all arguments to the call, as compared to functions in Python, which have a single implementation and no polymorphism (as opposed to Python method calls which use a different syntax and allows dispatch on the receiver of the method).
  • There are no classes in Julia. Instead there are structures (mutable or immutable), containing data but no methods.
  • Calling a method of a class instance in Python (x = MyClass(*args); x.f(y)) corresponds to a function call in Julia, e.g. x = MyType(args...); f(x, y). In general, multiple dispatch is more flexible and powerful than the Python class system.
  • Julia structures may have exactly one abstract supertype, whereas Python classes can inherit from one or more (abstract or concrete) superclasses.
  • The logical Julia program structure (Packages and Modules) is independent of the file structure (include for additional files), whereas the Python code structure is defined by directories (Packages) and files (Modules).
  • The ternary operator x > 0 ? 1 : -1 in Julia corresponds to a conditional expression in Python 1 if x > 0 else -1.
  • In Julia the @ symbol refers to a macro, whereas in Python it refers to a decorator.
  • Exception handling in Julia is done using trycatchfinally, instead of tryexceptfinally. In contrast to Python, it is not recommended to use exception handling as part of the normal workflow in Julia (compared with Python, Julia is faster at ordinary control flow but slower at exception-catching).
  • In Julia loops are fast, there is no need to write "vectorized" code for performance reasons.
  • Be careful with non-constant global variables in Julia, especially in tight loops. Since you can write close-to-metal code in Julia (unlike Python), the effect of globals can be drastic (see Performance Tips).
  • In Julia, rounding and truncation are explicit. Python's int(3.7) should be floor(Int, 3.7) or Int(floor(3.7)) and is distinguished from round(Int, 3.7). floor(x) and round(x) on their own return an integer value of the same type as x rather than always returning Int.
  • In Julia, parsing is explicit. Python's float("3.7") would be parse(Float64, "3.7") in Julia.
  • In Python, the majority of values can be used in logical contexts (e.g. if "a": means the following block is executed, and if "": means it is not). In Julia, you need explicit conversion to Bool (e.g. if "a" throws an exception). If you want to test for a non-empty string in Julia, you would explicitly write if !isempty(""). Perhaps surprisingly, in Python if "False" and bool("False") both evaluate to True (because "False" is a non-empty string); in Julia, parse(Bool, "false") returns false.
  • In Julia, a new local scope is introduced by most code blocks, including loops and trycatchfinally. Note that comprehensions (list, generator, etc.) introduce a new local scope both in Python and Julia, whereas if blocks do not introduce a new local scope in both languages.

Noteworthy differences from C/C++

  • Julia arrays are indexed with square brackets, and can have more than one dimension A[i,j]. This syntax is not just syntactic sugar for a reference to a pointer or address as in C/C++. See the manual entry about array construction.
  • In Julia, indexing of arrays, strings, etc. is 1-based not 0-based.
  • Julia arrays are not copied when assigned to another variable. After A = B, changing elements of B will modify A as well. Updating operators like += do not operate in-place, they are equivalent to A = A + B which rebinds the left-hand side to the result of the right-hand side expression.
  • Julia arrays are column major (Fortran ordered) whereas C/C++ arrays are row major ordered by default. To get optimal performance when looping over arrays, the order of the loops should be reversed in Julia relative to C/C++ (see relevant section of Performance Tips).
  • Julia values are not copied when assigned or passed to a function. If a function modifies an array, the changes will be visible in the caller.
  • In Julia, whitespace is significant, unlike C/C++, so care must be taken when adding/removing whitespace from a Julia program.
  • In Julia, literal numbers without a decimal point (such as 42) create signed integers, of type Int, but literals too large to fit in the machine word size will automatically be promoted to a larger size type, such as Int64 (if Int is Int32), Int128, or the arbitrarily large BigInt type. There are no numeric literal suffixes, such as L, LL, U, UL, ULL to indicate unsigned and/or signed vs. unsigned. Decimal literals are always signed, and hexadecimal literals (which start with 0x like C/C++), are unsigned, unless when they encode more than 128 bits, in which case they are of type BigInt. Hexadecimal literals also, unlike C/C++/Java and unlike decimal literals in Julia, have a type based on the length of the literal, including leading 0s. For example, 0x0 and 0x00 have type UInt8, 0x000 and 0x0000 have type UInt16, then literals with 5 to 8 hex digits have type UInt32, 9 to 16 hex digits type UInt64, 17 to 32 hex digits type UInt128, and more that 32 hex digits type BigInt. This needs to be taken into account when defining hexadecimal masks, for example ~0xf == 0xf0 is very different from ~0x000f == 0xfff0. 64 bit Float64 and 32 bit Float32 bit literals are expressed as 1.0 and 1.0f0 respectively. Floating point literals are rounded (and not promoted to the BigFloat type) if they can not be exactly represented. Floating point literals are closer in behavior to C/C++. Octal (prefixed with 0o) and binary (prefixed with 0b) literals are also treated as unsigned (or BigInt for more than 128 bits).
  • In Julia, the division operator / returns a floating point number when both operands are of integer type. To perform integer division, use div or ÷.
  • Indexing an Array with floating point types is generally an error in Julia. The Julia equivalent of the C expression a[i / 2] is a[i ÷ 2 + 1], where i is of integer type.
  • String literals can be delimited with either " or """, """ delimited literals can contain " characters without quoting it like "\"". String literals can have values of other variables or expressions interpolated into them, indicated by $variablename or $(expression), which evaluates the variable name or the expression in the context of the function.
  • // indicates a Rational number, and not a single-line comment (which is # in Julia)
  • #= indicates the start of a multiline comment, and =# ends it.
  • Functions in Julia return values from their last expression(s) or the return keyword. Multiple values can be returned from functions and assigned as tuples, e.g. (a, b) = myfunction() or a, b = myfunction(), instead of having to pass pointers to values as one would have to do in C/C++ (i.e. a = myfunction(&b).
  • Julia does not require the use of semicolons to end statements. The results of expressions are not automatically printed (except at the interactive prompt, i.e. the REPL), and lines of code do not need to end with semicolons. println or @printf can be used to print specific output. In the REPL, ; can be used to suppress output. ; also has a different meaning within [ ], something to watch out for. ; can be used to separate expressions on a single line, but are not strictly necessary in many cases, and are more an aid to readability.
  • In Julia, the operator (xor) performs the bitwise XOR operation, i.e. ^ in C/C++. Also, the bitwise operators do not have the same precedence as C/C++, so parenthesis may be required.
  • Julia's ^ is exponentiation (pow), not bitwise XOR as in C/C++ (use , or xor, in Julia)
  • Julia has two right-shift operators, >> and >>>. >> performs an arithmetic shift, >>> always performs a logical shift, unlike C/C++, where the meaning of >> depends on the type of the value being shifted.
  • Julia's -> creates an anonymous function, it does not access a member via a pointer.
  • Julia does not require parentheses when writing if statements or for/while loops: use for i in [1, 2, 3] instead of for (int i=1; i <= 3; i++) and if i == 1 instead of if (i == 1).
  • Julia does not treat the numbers 0 and 1 as Booleans. You cannot write if (1) in Julia, because if statements accept only booleans. Instead, you can write if true, if Bool(1), or if 1==1.
  • Julia uses end to denote the end of conditional blocks, like if, loop blocks, like while/ for, and functions. In lieu of the one-line if ( cond ) statement, Julia allows statements of the form if cond; statement; end, cond && statement and !cond || statement. Assignment statements in the latter two syntaxes must be explicitly wrapped in parentheses, e.g. cond && (x = value), because of the operator precedence.
  • Julia has no line continuation syntax: if, at the end of a line, the input so far is a complete expression, it is considered done; otherwise the input continues. One way to force an expression to continue is to wrap it in parentheses.
  • Julia macros operate on parsed expressions, rather than the text of the program, which allows them to perform sophisticated transformations of Julia code. Macro names start with the @ character, and have both a function-like syntax, @mymacro(arg1, arg2, arg3), and a statement-like syntax, @mymacro arg1 arg2 arg3. The forms are interchangeable; the function-like form is particularly useful if the macro appears within another expression, and is often clearest. The statement-like form is often used to annotate blocks, as in the distributed for construct: @distributed for i in 1:n; #= body =#; end. Where the end of the macro construct may be unclear, use the function-like form.
  • Julia has an enumeration type, expressed using the macro @enum(name, value1, value2, ...) For example: @enum(Fruit, banana=1, apple, pear)
  • By convention, functions that modify their arguments have a ! at the end of the name, for example push!.
  • In C++, by default, you have static dispatch, i.e. you need to annotate a function as virtual, in order to have dynamic dispatch. On the other hand, in Julia every method is "virtual" (although it's more general than that since methods are dispatched on every argument type, not only this, using the most-specific-declaration rule).

Noteworthy differences from Common Lisp

  • Julia uses 1-based indexing for arrays by default, and it can also handle arbitrary index offsets.

  • Functions and variables share the same namespace (“Lisp-1”).

  • There is a Pair type, but it is not meant to be used as a COMMON-LISP:CONS. Various iterable collections can be used interchangeably in most parts of the language (eg splatting, tuples, etc). Tuples are the closest to Common Lisp lists for short collections of heterogeneous elements. Use NamedTuples in place of alists. For larger collections of homogeneous types, Arrays and Dicts should be used.

  • The typical Julia workflow for prototyping also uses continuous manipulation of the image, implemented with the Revise.jl package.

  • For performance, Julia prefers that operations have type stability. Where Common Lisp abstracts away from the underlying machine operations, Julia cleaves closer to them. For example:

    • Integer division using / always returns a floating-point result, even if the computation is exact.
      • // always returns a rational result
      • ÷ always returns a (truncated) integer result
    • Bignums are supported, but conversion is not automatic; ordinary integers overflow.
    • Complex numbers are supported, but to get complex results, you need complex inputs.
    • There are multiple Complex and Rational types, with different component types.
  • Modules (namespaces) can be hierarchical. import and using have a dual role: they load the code and make it available in the namespace. import for only the module name is possible (roughly equivalent to ASDF:LOAD-OP). Slot names don't need to be exported separately. Global variables can't be assigned to from outside the module (except with eval(mod, :(var = val)) as an escape hatch).

  • Macros start with @, and are not as seamlessly integrated into the language as Common Lisp; consequently, macro usage is not as widespread as in the latter. A form of hygiene for macros is supported by the language. Because of the different surface syntax, there is no equivalent to COMMON-LISP:&BODY.

  • All functions are generic and use multiple dispatch. Argument lists don't have to follow the same template, which leads to a powerful idiom (see do). Optional and keyword arguments are handled differently. Method ambiguities are not resolved like in the Common Lisp Object System, necessitating the definition of a more specific method for the intersection.

  • Symbols do not belong to any package, and do not contain any values per se. M.var evaluates the symbol var in the module M.

  • A functional programming style is fully supported by the language, including closures, but isn't always the idiomatic solution for Julia. Some workarounds may be necessary for performance when modifying captured variables.