From 06b0081220d7e85134c61e6c86e723eb4a34b535 Mon Sep 17 00:00:00 2001
From: Matthew Almond <malmond@fb.com>
Date: Fri, 30 Apr 2021 13:11:04 -0700
Subject: [PATCH] Do not assume that a remote rpm is complete if present

We should not assume that a file present in the hashed directory for a remote
url is complete if present.

This manifests in two ways:

1. `Can not load RPM file: ` if the headers are incomplete
2. `Payload SHA256 digest: BAD` error otherwise

This isn't a common error in the field because most sites use `keepcache=0` in
`dnf.conf`. On the latter error the broken RPM is deleted, and the next run
will retry the download from scratch, and probably succeeding.

This impacts us because we use `keepcache=1` so once we get an interrupted
download, we can't (automatically) recover.

The fix here forces us to try to download/resume the file even if present on
disk, because we don't have any higher level metadata like digests to falsify
downloads without network.

This change makes an explicit decision to try to be more correct instead of
using an incomplete optimization.
---
 dnf/util.py | 2 --
 1 file changed, 2 deletions(-)

diff --git a/dnf/util.py b/dnf/util.py
index b9dcd47081..9f0eb4abd2 100644
--- a/dnf/util.py
+++ b/dnf/util.py
@@ -88,8 +88,6 @@ def _urlopen_progress(url, conf, progress=None):
     if progress is None:
         progress = dnf.callback.NullDownloadProgress()
     pload = dnf.repo.RemoteRPMPayload(url, conf, progress)
-    if os.path.exists(pload.local_path):
-        return pload.local_path
     est_remote_size = sum([pload.download_size])
     progress.start(1, est_remote_size)
     targets = [pload._librepo_target()]