disallow unrecognized string/char escapes

[stevengj]

As discussed in #21284, this PR disallows a backslash in string or character literals before unrecognized escape characters.

Besides the standard escapes that produce a different character (e.g. \n), the only allowed escapes where \* produces * are now for * ∈ {',",`,$,\}.

[stevengj]

(Note that I found a few bugs this way.)

[stevengj]

(Note that I found a few bugs this way.)

[stevengj]

I don't really understand the Travis failures. One appears to be a timeout, whereas the other is the mysterious:

	From worker 1:	compile: Test Failed
  Expression: Foo.g() === 97.0
   Evaluated: 2.0 === 97.0
Stacktrace:
 [1] (::Test62Main_compile.##1#14)() at /tmp/julia/share/julia/test/compile.jl:159

AppVeyor and FreeBSD and OSX Travis succeeded; is there some problem with Linux Travis now?

[stevengj]

I don't really understand the Travis failures. One appears to be a timeout, whereas the other is the mysterious:

	From worker 1:	compile: Test Failed
  Expression: Foo.g() === 97.0
   Evaluated: 2.0 === 97.0
Stacktrace:
 [1] (::Test62Main_compile.##1#14)() at /tmp/julia/share/julia/test/compile.jl:159

AppVeyor and FreeBSD and OSX Travis succeeded; is there some problem with Linux Travis now?

[vtjnash]

We've seen that one before. It seems to suggest we're losing a backedge at some point (stochastically), but we haven't reproduced it yet locally.

[vtjnash]

We've seen that one before. It seems to suggest we're losing a backedge at some point (stochastically), but we haven't reproduced it yet locally.

[JeffBezanson]

Travis is failing in all kinds of ways lately.

[JeffBezanson]

Travis is failing in all kinds of ways lately.

[StefanKarpinski]

The fact that this revealed a bunch of bugs is a pretty good indicator that this was a good idea.

[StefanKarpinski]

The fact that this revealed a bunch of bugs is a pretty good indicator that this was a good idea.

[stevengj]

There was a suggestion on discourse that this should be changed to a warning.

On the one hand, it seems like virtually all the errors that are popping up are bugs. On the other hand, we normally make changes like this a warning for a while...

[stevengj]

There was a suggestion on discourse that this should be changed to a warning.

On the one hand, it seems like virtually all the errors that are popping up are bugs. On the other hand, we normally make changes like this a warning for a while...

[JeffBezanson]

These are very easy to fix, and the fixed code is both backwards and forwards compatible, so it doesn't seem so crucial to me to give a warning.

[JeffBezanson]

These are very easy to fix, and the fixed code is both backwards and forwards compatible, so it doesn't seem so crucial to me to give a warning.

[StefanKarpinski]

Especially since when someone upgrades to 0.7, they should fix all warnings and then upgrade to 1.0, at which point this would fail anyway. There's even a case to be made that 0.7 should have depwarn=error on by default, but that's a different discussion.

[StefanKarpinski]

Especially since when someone upgrades to 0.7, they should fix all warnings and then upgrade to 1.0, at which point this would fail anyway. There's even a case to be made that 0.7 should have depwarn=error on by default, but that's a different discussion.

[ChrisRackauckas]

Did this cause math blocks to no longer work?

"""
    calculate_residuals!(out, ũ, u₀, u₁, α, ρ)

Save element-wise residuals
```math
\frac{ũ}{α+\max{|u₀|,|u₁|}*ρ}
```
in `out`.
"""
function calculate_residuals!(out, ũ, u₀, u₁, α, ρ, internalnorm)
   @. out = ũ / (α + max(internalnorm(u₀), internalnorm(u₁)) * ρ)
end

ERROR: syntax: invalid escape sequence

[ChrisRackauckas]

Did this cause math blocks to no longer work?

"""
    calculate_residuals!(out, ũ, u₀, u₁, α, ρ)

Save element-wise residuals
```math
\frac{ũ}{α+\max{|u₀|,|u₁|}*ρ}
```
in `out`.
"""
function calculate_residuals!(out, ũ, u₀, u₁, α, ρ, internalnorm)
   @. out = ũ / (α + max(internalnorm(u₀), internalnorm(u₁)) * ρ)
end

ERROR: syntax: invalid escape sequence

[KristofferC]

Just escape the backslash?

[KristofferC]

Just escape the backslash?

[ChrisRackauckas]

Having to go and escape the TeX code in every docstring is a pretty heavy handed workaround. For normal strings you'd just use LaTeXStrings.jl. Is there no simple way to write TeX in docstrings now? (Do string macros work on docstrings? I guess I never checked)

[ChrisRackauckas]

Having to go and escape the TeX code in every docstring is a pretty heavy handed workaround. For normal strings you'd just use LaTeXStrings.jl. Is there no simple way to write TeX in docstrings now? (Do string macros work on docstrings? I guess I never checked)

[stevengj]

Note that \frac didn't work previously, either; it would have given frac since \f was equivalent to f. So, throwing an error instead caught a bug.

However, you can just use doc"..." to avoid the need for backslash escapes.

doc"""
    calculate_residuals!(out, ũ, u₀, u₁, α, ρ)

Save element-wise residuals
```math
\frac{ũ}{α+\max{|u₀|,|u₁|}*ρ}
```
in `out`.
"""

[stevengj]

Note that \frac didn't work previously, either; it would have given frac since \f was equivalent to f. So, throwing an error instead caught a bug.

However, you can just use doc"..." to avoid the need for backslash escapes.

doc"""
    calculate_residuals!(out, ũ, u₀, u₁, α, ρ)

Save element-wise residuals
```math
\frac{ũ}{α+\max{|u₀|,|u₁|}*ρ}
```
in `out`.
"""

[stevengj]

The manual has a brief comment about this:

@doc_str should only be used when the docstring contains $ or \ characters that should not be parsed by Julia such as LaTeX syntax or Julia source code examples containing interpolation.

[stevengj]

The manual has a brief comment about this:

@doc_str should only be used when the docstring contains $ or \ characters that should not be parsed by Julia such as LaTeX syntax or Julia source code examples containing interpolation.