Yep, I think using relative jumps would fix a lot of these issues, but
we would have to do a survey of where jump and split instructions are
used, to make sure we catch all scenarios where something could be
inserted in the middle, as you say. My main worry is that relative jumps
might be more difficult to reason about.

I've applied this patch series, but see: https://lists.sr.ht/~sircmpwn/hare-dev/%3C0d1adb8a-1842-4364-881e-d7163d919d0b@app.fastmail.com%3E

This bug has now been fixed. For more details, see: https://lists.sr.ht/~sircmpwn/hare-dev/%3C0d1adb8a-1842-4364-881e-d7163d919d0b@app.fastmail.com%3E

I just pushed some commits to master with some regex fixes. I would have
sent a patch series, but for various reasons you'll see below, it would
have gotten too confusing, and the changes are already confusing enough,
so I'm just documenting what I did here.

First of all, Sebastian implemented nested capture groups:
https://lists.sr.ht/~sircmpwn/hare-dev/patches/56226

That patch had an unnecessary dependency on the nomem patch, but I
modified it to remove the dependency and I applied it.

But this patch also uncovered a different issue.

*The Shifting Issue*

Thanks! There's already a patch that fixes this. I'm just waiting for an update from Sebastian (cc'd).

Signed-off-by: Vlad-Stefan Harbuz <vlad@vlad.website>
---

This addresses a discussion with bgs2 on IRC.

v1 -> v2:
* made the new test fail if the inst_split offset is incorrectly
    `len(insts) + 1`

 regex/+test.ha | 2 ++
 regex/regex.ha | 4 ++++
 2 files changed, 6 insertions(+)

diff --git a/regex/+test.ha b/regex/+test.ha
[message trimmed]

Signed-off-by: Vlad-Stefan Harbuz <vlad@vlad.website>
---
 regex/+test.ha | 2 ++
 regex/regex.ha | 4 ++++
 2 files changed, 6 insertions(+)

diff --git a/regex/+test.ha b/regex/+test.ha
index 279eeea7..40097b2c 100644
--- a/regex/+test.ha
+++ b/regex/+test.ha
@@ -412,6 +412,8 @@ fn run_rawreplace_case(
 		(`^x(abc){,2}$`, "xabc", matchres::MATCH, 0, -1),
 		(`^x(abc){,2}$`, "xabcabc", matchres::MATCH, 0, -1),
 		(`^x(abc){,2}`, "xabcabcabc", matchres::MATCH, 0, 7),
[message trimmed]

Gotcha. Could we split the shift() stuff out into a different (prior) commit, so that I can test it separately? Tysm :)

Done — would recommend `git rebase upstream/main`.

Oh yeah there's some missing stuff — I'll just pull it in. Thanks!