`M-x pdf-sync-forward-search` misplaces arrow.

[YueRen]

I believe this may be a bug with pdf-tools, but feel free to close the issue if it is not.

Describe the bug
When using auctex in conjunction with pdf-tools, the command M-x pdf-sync-forward-search (used in M-x TeX-view bound to C-c C-v) places an arrow indicating in the pdf buffer the position where the point is in the tex buffer.

In the past, if the pdf is larger than the window, the arrow is still placed within the pdf buffer (or on the border, cannot remember clearly sry). Since a few days ago, the placement of the arrow starting ignoring the window boundaries. Instead, the arrow seems to be placed where the correct position would be, if the window boundaries would not exist, even if this is in the middle of another window.

Here a screenshot showing the arrow placement when the pdf fits the window:

Here a screenshot showing the arrow placement when the pdf extends beyond the window:

Here is the init.el used in the screenshots above:

(require 'package)
(setq package-archives '(("elpa" . "https://elpa.gnu.org/packages/")
                         ("melpa" . "https://melpa.org/packages/")
                         ("org" . "https://orgmode.org/elpa/")))
(package-initialize)
(unless package-archive-contents (package-refresh-contents))
(require 'use-package-ensure)
(setopt use-package-always-ensure t)

(use-package pdf-tools
  :mode "\\.pdf\\'"
  :init
  (pdf-loader-install))

(use-package tex
  :ensure auctex
  :mode
  ("\\.tex\\'" . latex-mode)
  :custom
  (TeX-view-program-selection '((output-pdf "PDF Tools")))
  (TeX-view-program-list '(("PDF Tools" TeX-pdf-tools-sync-view)))
  (TeX-source-correlate-mode t)
  (TeX-source-correlate-start-server t)
  :init
  (add-hook 'TeX-after-compilation-finished-functions #'TeX-revert-document-buffer))

Here is mwe.tex:

\documentclass{article}
\usepackage{lipsum} % Provides the \lipsum command

\begin{document}

\section{Introduction}
\lipsum[1] % Generates the 1st paragraph of Lorem ipsum

\section{Main Content}
\lipsum[2-4] % Generates paragraphs 2 through 4

\end{document}

%%% Local Variables:
%%% mode: LaTeX
%%% TeX-master: t
%%% End:

Steps to Reproduce the behaviour

1. Open mwe.tex
2. Compile mwe.tex using C-c C-c (repeat if necessary)
3. Run TeX-view via C-c C-v

Desktop
My laptop uses Ubuntu 25.10 and have Emacs installed via snap

GNU Emacs 30.2
Development version ff7d3f02b943 on master branch; build date 2025-12-17.
Copyright (C) 2025 Free Software Foundation, Inc.
GNU Emacs comes with ABSOLUTELY NO WARRANTY.
You may redistribute copies of GNU Emacs
under the terms of the GNU General Public License.
For more information about these matters, see the file named COPYING.

Your pdf-tools install
This is my pdf-tools version:

     Status: Installed in ‘pdf-tools-20260102.1101/’ (unsigned).
    Version: 20260102.1101
     Commit: e4b7f1f37cf59ddf025d609ffcdabe732a6e99ba
    Summary: Support library for PDF documents.
   Requires: emacs-26.3, tablist-1.0, let-alist-1.0.4
Required by: org-pdftools-20250714.1631
    Website: http://github.com/vedang/pdf-tools/
   Keywords: files multimedia 
     Author: Andreas Politz <mail@andreas-politz.de>
Other versions: 20260102.1101 (melpa).

[aikrahguzar]

Most likely this has been introduced by me while trying to integrate this with pdf-view-roll-minor-mode. I will try to find the cause next weekend or so. In the meantime, if possible can you please try my fork and see if this issue has been fixed there? It also has more accurate forward search and the option to highlight the word instead of (or in addition to) placing the tooltip arrow.

[YueRen]

The issue seems to persist in your fork. Not sure, why but I also had to add a (require 'reftex) to get the syncing to work (could have been missing from my initially posted init.el):

(require 'package)
(setq package-archives '(("elpa" . "https://elpa.gnu.org/packages/")
                         ("melpa" . "https://melpa.org/packages/")))
(package-initialize)
(unless package-archive-contents (package-refresh-contents))
(require 'use-package-ensure)
(setopt use-package-always-ensure t)

(use-package pdf-tools
  :mode "\\.pdf\\'"
  :vc ( :url "https://codeberg.org/rahguzar/pdf-tools.git"
        :lisp-dir "lisp"
        :rev :newest)
  :init
  (pdf-loader-install))

(use-package tex
  :ensure auctex
  :mode
  ("\\.tex\\'" . latex-mode)
  :custom
  (TeX-view-program-selection '((output-pdf "PDF Tools")))
  (TeX-view-program-list '(("PDF Tools" TeX-pdf-tools-sync-view)))
  (TeX-source-correlate-mode t)
  (TeX-source-correlate-start-server t)
  :init
  (add-hook 'TeX-after-compilation-finished-functions #'TeX-revert-document-buffer))
(require 'reftex)

[aikrahguzar]

The need for (require 'reftex) is an omission due to me and pretty easy to fix.

I think I understand the reason for the original error too. I will try to get them both fixed soonish. I think in the meantime pdf-auto-slice-minor-mode will reduce the need for nonzero hscroll which is what is causing the bug.

[aikrahguzar]

I pushed a commit to my fork that I think should fix this issue. It is a bit involved but the two users of tooltips (links and sync) seem to be working fine. Please test and see if it improves the situation for you.

[YueRen]

Seems to be working fine (in fact, better than before)! Thank you!

[aikrahguzar]

Have you tried the new highlighting mode for forward sync? Any opinion/suggestion about it?

[YueRen]

I didn't even have to try it, it is enabled by default (with the config I posted above).

It's quite nice, perfectly accurate with text in my testing, but less accurate with math symbols when latex commands are involved (e.g., \rightarrow). It always highlights the right equation so it's not a big deal, but the element it highlights is often off.

Personal opinion: Latex commands can hide all sorts of stuff behind them and I wouldn't be surprised if highlighting math elements accurately is nigh impossible because of that. If that is the case, I would probably prefer just highlighting the entire formula instead of trying to guess which element the point is on.

[aikrahguzar]

I didn't even have to try it, it is enabled by default (with the config I posted above).

I thought I had conditioned it on the availability of synctex executable but I see that I haven't. My main worry is that the more accurate highlighting mode will be quite slow for people on an older machine. But personally I don't feel much of a difference between pdf-sync-forward-use-heuristic being nil or t.

It's quite nice, perfectly accurate with text in my testing, but less accurate with math symbols when latex commands are involved (e.g., \rightarrow). It always highlights the right equation so it's not a big deal, but the element it highlights is often off.

Both forward and reverse search work better with text than math. That is to be expected since the heuristic used it the alignment of words in source and text. The alignment of bigger words matter more than that of smaller words. In text there are usually big words not far from the cursor, in formulas there are basically only small "words". In my experience although the highlight is often off in the math mode it is only off by a small amount.

There are some cases in which we can't refine what synctex gives us so we just end up with a big highlighted regions. It is pretty rare though and for me mostly happens with stuff like \maketitle.

Personal opinion: Latex commands can hide all sorts of stuff behind them and I wouldn't be surprised if highlighting math elements accurately is nigh impossible because of that. If that is the case, I would probably prefer just highlighting the entire formula instead of trying to guess which element the point is on.

You are correct that it is almost impossible. That it works as well as it does is due to contents of the pdf-sync-backward-text-translations. If you want to try line level highlighting, put this in your config:

(advice-add 'pdf-sync--forward-highlight :override
            (defun +pdf-sync--forward-highlight (edges page)
              "Temporarily highlight EDGES on PAGE."
              (when pdf-sync--forward-timer (cancel-timer pdf-sync--forward-timer))
              (setq-local pdf-sync--forward-timer
                          (run-with-timer 3 nil #'pdf-sync--forward-redisplay
                                          (current-buffer) (selected-window)))
              (pdf-view-display-region `(,page ,edges) nil 'line)))

It might works better for you. This is not highlighting the entire formula because I don't know how to determine where a formula starts or ends. You can copy the contents of a page in a pdf buffer and paste it somewhere. What we do is to compare this text against the text in the tex buffer as best we can.