Javascript (Firefox), 29 27 bytes

new Date('-').toISOString() 

throws RangeError: invalid date which is 24 bytes. Tested on Firefox 54.0.1 on Windows 10.

Python 3, 29 Bytes

from __future__ import braces 

Error

SyntaxError: not a chance 

Aubergine, 53 bytes

0/0 Lots of wasted space. I don't know how Aubergine. 

Outputs SyntaxError: Invalid instruction (0) at character 0.

Try it online!

GW-BASIC, 13 bytes

------------- 

Error (12 bytes):

Syntax error 

As GW-BASIC treats anything it doesn't recognize as a syntax error, there are a near-infinite amount of strings of length 13 that I could've used instead of -------------

Ruby, 25 bytes

That was a fun exercise, thanks! There's probably a way to get a shorter error message with a SegFault, but I couldn't find any.

/\x # Invalid hexa regexp 

Error message:

a:1: invalid hex escape 

24 bytes, including a trailing newline.

Try it online!

Ruby, 26 bytes

Here's my previous answer:

08 # No 8 allowed in octal 

Error message:

a:1: Invalid octal digit 

25 bytes, including a trailing newline.

Bash, 29 bytes

true&&false||false||a 

error message:

a: command not found 

it's really just filler of bash builtins before using an undefined command. This was the shortest bash error message I could think of.

TRS-80 Model 100 Basic, 10 bytes

Enter this:

?????????? 

Resulting error message:

?SN Error 

All Basic error codes on this charming little machine are retrieved by using the error code to index into an array in ROM that looks like NFSNRG... and so on, so all error codes must be exactly 2 characters. With the 7 extra characters taken into account, all error messages will always be 9 bytes and hence 10 bytes is the shortest possible on this machine.

This is just one way to do it; there are countless more. It generates error 2, code SN, which means syntax error.

C++ (on macOS High Sierra beta), 23 characters

int main(){*(int*)0=0;} 

Output: (22 characters)

Segmentation fault: 11 

I do not recall whether previous versions of macOS provide the more traditional Segmentation fault (core dumped) response but adding spaces to the code to pad that out is pretty trivial.

LibreOffice Calc, 7 bytes

=aaaaaa 

Result:

#NAME? 

C#, 46 bytes

class P{public static int Main(string[]arg){}} 

And produces the error at 45 bytes:

'P.Main()': not all code paths return a value

Probably a shorter way to do this in C# but I can't think of it.

Java 7, 43 bytes

Based on a comment from the OP, we are allowed to assume the filename is 1 character long. This program has that assumption and my example uses the source code's filename z.java

// public static void main(String[] args){} 

Error message:

Error: Could not find or load main class z 

R, 9 bytes

[edited, because I can't count to 8]

code (9 bytes):

stop();;; 

output (8 bytes, including trailing space and newline):

Error: 

According to the manual, "stop stops execution of the current expression and executes an error action." The first argument is meant to be an error message, but can be omitted.

dc (GNU 1.2) on Windows 10 x64, 185 bytes

dc doesn't crash fatally very often, and I couldn't be bothered to remember code that causes a segfault. I did recall some weird behavior involving arrays that causes a sad and total loss of the current session.

 zzz zzzzzzz zz zzzzz zzzzz zzzzzzzzzz zzzz zzzzzzzzzzz zzz zzzzzzzzz zzz zzzzzzz zz zzzzzzzzz zz zz zz zzzzzzz zzzz zzzzzz zzzzzzz zzz zzzzzzzzzzzzz zzzzzzz zzzz zzz zzzz zzzzzzzz:zlz 

Pushes numbers on the stack until the last five characters. :z stores the second-to-topmost number in the topmost-indexed slot in the array named z. Typically, this array would be linked through time and space to the current instance in the register named z. However, we haven't actually put anything there. Next, we try to lz, or copy the value on top of z to the top of the main stack. We can't: there's nothing there. I'm not sure why this mechanism triggers the failure that it does, but it works. (Or...doesn't.) The trailing newline is significant for two reasons: it puts the code in the black at 185, and pasting it into dc causes it to crash immediately.

Basically, we're running across a canyon, and as long as we don't look down, we're good. But once we look and realize we're floating in mid-air, we fall and crash hard, and little stars zoom around our poor befuddled heads.

Error, 184 bytes

 dc: garbage in value being duplicated This application has requested the Runtime to terminate it in an unusual way. Please contact the application's support team for more information. 

I'm not sure if only part of that is the actual error or what, so feel free to correct me. The code is obviously very flexible when it comes to length.

TI-Basic, 9 bytes

Shortest error messages are 8 bytes each: ERR:DATE, ERR:MODE, ERR:STAT, and ERR:ZOOM. I didn't consider ERR:DATE because that doesn't work on models without an internal clock. Also, I didn't go for ERR:ZOOM because it seemed too hard to trigger.

Program (9 bytes):

Seq:DrawInv X::::: 

Error message: ERR:MODE (8 bytes)

Program (9 bytes):

median({1},{0:: 

Error message: ERR:STAT (8 bytes)

Shakespeare Programming Language, 48 bytes

,a.Ajax,a.Act I:a.Scene I:a.[Enter Ajax]I error! 

Try it online!

Produces:

Unrecognized error encountered. No code output. 

There's not much we can do about creating runtime errors, because the minimum overhead to create a program with two characters is 56 bytes, and you can't do much with one character. Instead, we settle for confusing the compiler, which means finding a sentence structure which doesn't fit the grammar rules. In this case, anything that isn't a character name after a character enters the stage.

Shorter error messages exist, for example Error at line 1: act expected, but the code compiled isn't valid, so you get an extra few lines of gibberish after the first error message.

Sinclair ZX81 - 8 bytes 4 bytes 2 bytes (2 BASIC tokens) using direct mode:

Newest solution

 PRINT A 

Less old solution

 PRINT A+A 

Old solution (8 bytes of memory):

 1 PRINT A 

When you enter the command RUN, the following error is reported:

2/1 

as in the screen shot below - Error code 2 means "Undefined variable" or something similar.

IE9 Chinese version, 9 bytes

'12345'() 

Outputs (assuming GBK encoding)

缺少函数 

Vim, 17 bytes

 :w { 

Try it online!

When you write to a file and you have an unmatched starting curly brace in the filename, it will give the error E220: Missing }. The code to get this error is very short, so we just pad it out with spaces to get to the length we want. You can't see this error on TIO, but it works in the terminal.

CPython 3.9, 71 bytes

__peg_parser__ ######################################################## 

Not the shortest, but I found this "feature" worth sharing nonetheless: when Python 3.9 switched to a PEG parser, there was a special easter egg keyword __peg_parser__ added which immediately causes a syntax error. The error outputted is:

 File "a", line 1 __peg_parser__ ^ SyntaxError: You found it! 

This was removed in Python 3.10.

/bin/sh ("dash" on Ubuntu Bionic), 20 bytes

cd /;ok="okokokok";@ 

Output is:

sh: 1: @: not found 

The cd / command it just to make sure to switch to a location where there isn't a file named @ since that's effectively what causes the error.

Vyxal, 3 bytes

1ƛ° 

Try it Online! -1 byte thanks to pacman256.

Outputs ⟨ .

Vyxal's online mode has a weird quirk where, if an error occurs inside a lazily-evaluated list, the error message won't actually get printed. We're not entirely sure why this happens, but this means the entirety of stdout is Vyxal starting to print a list ⟨ , which I think can reasonably be considered an error message - especially since, outside of explicitly printing that string, it'd never occur in normal use.

Aside from that, it's actually kinda difficult to throw a runtime error in Vyxal. Basically every single-byte builtin has either type coercion or useful behaviour for all types, so a previous version of this used the two-byte run-length decoding builtin ød which expects a list.

However, this version uses °, Vyxal's almost-never-used complex number format. x°y translates to x + iy, and the cases x° and °y become x + i and iy respectively. But ° alone throws a syntax error.

Except, it doesn't throw an error at parse time, nor does it throw a syntax error when the transpiled Python is evaluated. Rather, it's compiled to sympy.nsimplify("+*I"), where there's nothing due to the lack of either real parts or imaginary parts - and that throws an error.

Brain-Flak, 46 bytes

)))))))))))))))))))))))))))))))))))))))))))))) 

Try it online!

This prints

Error at character 0: Unopened ')' character. 

because in Brain-flak, all brackets must be balanced for the program to be valid.

J, 10 bytes

0000000^.0 

NaN error <-------- length 9

Japt, 32 bytes

???????????????????????????????? 

Error message (31 bytes):

SyntaxError: Unexpected token ?

Try it online!

Rexx (Regina), 25 bytes

interpret interpret 2+3+1 

Try it online!

"interpret" treats the text following it as source and tries to execute it. Since 2+3+1 is not included in quotes it treats it as a calculation and does that first. Resulting in "interpret interpret 6" The first "interpret" tells REXX to treat "interpret 6" as source. So it executes it. This results in the attempt to execute 6 as source.

sh: 6: command not found 

JAVA 8, 137 bytes error and 138 bytes code

package a; public class a{ public static void main(String[] args){ String f = "sdfgsfdgdffffffffffffffffffffffffsfdgfd"; a = 5; } } 

produces the error:

Exception in thread "main" java.lang.Error: Unresolved compilation problem: a cannot be resolved to a variable at a.a.main(a.java:6) 

as this is my first time please tell me if i did something wrong

Batch 90 18 bytes

@set/ab=1+*2*3*4*5 

Missing Operand\ Golfed by SteveFest

Eukleides, 17 bytes

Program:

a=centroid(empty) 

Error:

X:1: empty set. 

(Plus a newline). All fatal errors in Eukleides begin with (filename):(line number): and end with a period and a newline. I combed the source, and found that empty set was the shortest error. At first, I thought I was out of luck, because I only saw the lengthy command isobarycenter throwing it, but spotted centroid throwing it as well. Then I couldn't figure out how to make an empty set; all my attempts got me a syntax error first. Turns out, empty is a constant for an empty set, something I never needed to know before. I am quite confident this is the shortest Eukleides solution.

PHP, 56 bytes

<?php aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa 

produces

PHP Parse error: syntax error, unexpected end of file 

Octave, 28 bytes

a=[];'Will cause error';a(1) 

Try it online!

This throws this error:

error: a(1): out of bound 0 

Damn, golfing error messages was cumbersome! I have rewritten this answer 8 times now, since I've gradually found error messages that were one byte shorter than the previous one. I think this answer took me an hour, since I've made a complete rewrite 8 times... :(

Why is this the shortest error message?

• Relying on undefined functions or variables is not short enough (41)

   error: 'a' undefined near line 1 column 1 

• Relying on the wrong input type is not short enough (38)

  error: mod: wrong type argument 'sq_string' error: whos: all arguments must be strings error: sum: wrong type argument 'cell' 

• Relying on syntax errors is definitely not short enough (42)

  parse error: syntax error >>> & ^ 

• Wrong indexing is not short enough (38)

  error: scalar cannot be indexed with { 

• Short function names and inputs that cause errors is not short enough (34)

  error: a: No such file or directory error: load: unable to find file a 

I've looked through all the functions in Octave, and I can't imagine anything being shorter than this.