talk: use tcolorbox + minted

Drop minted hack to fix box alignment. Was broken with recent TeX stacks
(texlive 2023.20231207).

talk/main.tex

@@ -9,7 +9,7 @@
 \usepackage{bm}
 \usepackage{xspace}
 \usepackage{amsmath}
-\usepackage{minted}
+\usepackage{tcolorbox}
 \usepackage[
     backend=biber,
     maxbibnames=2,
@@ -26,30 +26,35 @@
 % minted
 %----------------------------------------------------------------------------------%
 
-% https://tex.stackexchange.com/a/173854
-%
-%% fix the minted@colorbg environment
-\makeatletter
-\renewenvironment{minted@colorbg}[1]
- {\def\minted@bgcol{#1}%
-  \noindent
-  \begin{lrbox}{\minted@bgbox}
-  \begin{minipage}{\linewidth-2\fboxsep}}
- {\end{minipage}%
-  \end{lrbox}%
-  \setlength{\topsep}{\bigskipamount}% set the vertical space
-  \trivlist\item\relax % ensure going to a new line
-  \colorbox{\minted@bgcol}{\usebox{\minted@bgbox}}%
-  \endtrivlist % close the trivlist
- }
-\makeatother
-
-\definecolor{mintedbg}{rgb}{0.95,0.95,0.95}
-\setminted{
-    bgcolor=mintedbg,
-    fontsize=\small,
+
+\tcbuselibrary{minted}
+
+\tcbset{%
+    % must use \tcbuselibrary{minted}
+    %%listing engine=minted,
+    minted options={fontsize=\small},
+    minted style=gruvbox-light,
+    listing only,
+    boxsep=0pt,
+    leftrule=0pt,
+    rightrule=0pt,
+    toprule=0pt,
+    bottomrule=0pt,
+    right=2pt,
+    top=2pt,
+    bottom=2pt,
+}
+
+\newtcbinputlisting[]{\mintedfromfile}[2]{%
+    minted language=#1,
+    listing file={#2},
+    left=20pt,
 }
 
+\newtcblisting{mintedcode}[1]{%
+    minted language=#1,
+    left=0pt,
+}
 
 %----------------------------------------------------------------------------------%
 % beamer style and layout
@@ -178,36 +183,36 @@
         f(x) = \ln(\sin^2(x))
     \end{equation*}
     Code:
-    \begin{minted}{python}
+    \begin{mintedcode}{python}
     f = lambda x: np.log(np.power(np.sin(x), 2))
-    \end{minted}
+    \end{mintedcode}
     Derivative:
     \begin{equation*}
         \td{f}{x} = \red{\td{c}{b}}\,\green{\td{b}{a}}\,\blue{\td{a}{x}}
             = \red{\frac{1}{\sin^2(x)}}\,\green{2\,\sin(x)}\,\blue{\cos(x)}
             = 2\,\frac{\cos(x)}{\sin(x)} = 2\,\cot(x)
     \end{equation*}
     Code:
-    \begin{minted}{python}
+    \begin{mintedcode}{python}
     fprime = lambda x: 2 * np.cos(x) / np.sin(x)
-    \end{minted}
+    \end{mintedcode}
 \end{frame}
 
 
 \begin{frame}
     \frametitle{AD teaser: arbitrary code}
     \only<1>{
-        \inputminted{python}{code/jax_ad_teaser_func.py}
+        \mintedfromfile{python}{code/jax_ad_teaser_func.py}
         \vfill
     }
     \only<2->{
         \uncover<2->{
-            \inputminted{python}{code/jax_ad_teaser_func_w_imports.py}
+            \mintedfromfile{python}{code/jax_ad_teaser_func_w_imports.py}
         }
     }
     \vspace{-0.3cm}
     \uncover<3->{
-        \inputminted{python}{code/jax_ad_teaser_grad_usage.py}
+        \mintedfromfile{python}{code/jax_ad_teaser_grad_usage.py}
     }
 \end{frame}
 
@@ -221,9 +226,9 @@
         \ve f(\ve x) = \ve c(\ve b(\ve a(\ve x)))
     \end{equation*}
     For $n=m$:
-    \begin{minted}{python}
+    \begin{mintedcode}{python}
     f = lambda x: np.log(np.power(np.sin(x), 2))
-    \end{minted}
+    \end{mintedcode}
     \vfill
     \uncover<2->{
         Jacobian:
@@ -313,26 +318,26 @@
 \begin{frame}
     \frametitle{Custom JVPs in \jax}
     \only<1>{
-        \inputminted{python}{code/jax_defjvp_mysin_no_decorator.py}
+        \mintedfromfile{python}{code/jax_defjvp_mysin_no_decorator.py}
     }
     \only<2->{
         \uncover<2->{
-            \inputminted{python}{code/jax_defjvp_mysin.py}
+            \mintedfromfile{python}{code/jax_defjvp_mysin.py}
         }
     }
     \vspace{-0.3cm}
     \uncover<3->{%
-        \inputminted{python}{code/jax_defjvp_with_jac.py}
+        \mintedfromfile{python}{code/jax_defjvp_with_jac.py}
     }
     \vspace{-0.3cm}
     \uncover<4->{%
-        \inputminted{python}{code/jax_defjvp.py}
+        \mintedfromfile{python}{code/jax_defjvp.py}
     }
 \end{frame}
 
 \begin{frame}
     \frametitle{Default JVPs for \numpy primitives in \jax}
-    \inputminted{python}{code/jax_lax_sin.py}
+    \mintedfromfile{python}{code/jax_lax_sin.py}
 \end{frame}
 
 \begin{frame}
@@ -382,8 +387,8 @@
 
 \begin{frame}
     \frametitle{Reverse mode VJPs in \pytorch}
-    \url{https://github.com/pytorch/pytorch/blob/master/tools/autograd/derivatives.yaml}
-    \inputminted{yaml}{code/pytorch_vjp.yaml}
+    \url{https://github.com/pytorch/pytorch/blob/main/tools/autograd/derivatives.yaml}
+    \mintedfromfile{yaml}{code/pytorch_vjp.yaml}
     Scripts to generate C++ code (\co{libtorch.so}) and Python bindings.
 \end{frame}
 
@@ -397,23 +402,24 @@
         \begin{equation*}
             f(\ve x) = c(\ve b(\ve a(\ve x)))
         \end{equation*}
-        \inputminted{python}{code/scalar_field.py}
+        \mintedfromfile{python}{code/scalar_field.py}
     }
 \end{frame}
 
-\begin{frame}[fragile]
+%%\begin{frame}[fragile]
+\begin{frame}
     \frametitle{\pytorch: Reverse mode example}
     \begin{equation*}
         f: \mathbb R^n\ra \mathbb R, \quad f(\ve x) = c(\ve b(\ve a(\ve x)))
     \end{equation*}
-    \inputminted[xleftmargin=.5cm]{python}{code/pytorch_fwd_rev_1.py}
+    \mintedfromfile{python}{code/pytorch_fwd_rev_1.py}
     \uncover<2->{
         Forward pass, tracing, observe \texttt{grad\textunderscore fn}
-        \inputminted[xleftmargin=.5cm]{python}{code/pytorch_fwd_rev_2.py}
+        \mintedfromfile{python}{code/pytorch_fwd_rev_2.py}
     }
     \uncover<3->{
         Backward pass
-        \inputminted[xleftmargin=.5cm]{python}{code/pytorch_fwd_rev_3.py}
+        \mintedfromfile{python}{code/pytorch_fwd_rev_3.py}
     }
 \end{frame}
 
@@ -430,26 +436,26 @@
     \end{align*}
     \uncover<2->{
         Default: initialize backward pass with $\pdi{f}{f}=1$
-        \inputminted[xleftmargin=.5cm]{python}{code/pytorch_rev_detail_1.py}
+        \mintedfromfile{python}{code/pytorch_rev_detail_1.py}
     }
     \uncover<3->{
         VJP: extract one row of Jacobian $\pdi{\ve b}{\ve a}$
-        \inputminted[xleftmargin=.5cm]{python}{code/pytorch_rev_detail_2.py}
+        \mintedfromfile{python}{code/pytorch_rev_detail_2.py}
     }
 \end{frame}
 
 \begin{frame}[fragile]
     \frametitle{Higher order fun}
-    \inputminted{python}{code/higher_order_1.py}
+    \mintedfromfile{python}{code/higher_order_1.py}
     \vspace{-0.3cm}
     \uncover<2->{
-        \inputminted{python}{code/higher_order_2.py}
+        \mintedfromfile{python}{code/higher_order_2.py}
     }
     \uncover<3->{
         $\ma J:=\partial f; \ma H = \partial^2 f = \partial(\partial f) = \partial(\nabla f)$ for $f:\mathbb R^n\ra\mathbb R$
     }
     \uncover<4->{
-        \inputminted{python}{code/higher_order_hessian.py}
+        \mintedfromfile{python}{code/higher_order_hessian.py}
     }
 \end{frame}
 
@@ -459,15 +465,15 @@
     \begin{itemize}
         \item restricted to \numpy/\pytorch-based code or need to wrap target package
         \item in-place ops (\co{a[i] *= 3}) can break AD, \pytorch: Tensor versioning system
-        \begin{minted}{python}
+        \begin{mintedcode}{python}
     RuntimeError: one of the variables needed for gradient
     computation has been modified by an inplace operation:
     [torch.FloatTensor []], which is output 0 of
     SelectBackward, is at version 55; expected version 51
     instead. Hint: enable anomaly detection to find the
     operation that failed to compute its gradient, with
     torch.autograd.set_detect_anomaly(True).
-        \end{minted}
+        \end{mintedcode}
         \item numerical accuracy a.k.a. when do you want \verb|custom_jvp|
         \begin{itemize}
             \item AD derivatives can generate numerically unstable code
@@ -485,7 +491,7 @@
 
 \begin{frame}
     \frametitle{\texttt{jaxpr}}
-    \inputminted{python}{code/jaxpr.py}
+    \mintedfromfile{python}{code/jaxpr.py}
 \end{frame}
 
 \end{document}