You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by Ivan Zhakov <ch...@gmail.com> on 2005/09/26 14:52:50 UTC

wc-replacements: text base checksums

Hi!
I studying working copy checksums for text base and revert base
behavior. I wonder that checksums is optional for text base. For
example if we somewhere forget write checksum for text base, working
copy would silently apply and commit txt-delta without checking
checksum:
subversion/libsvn_wc/update_editor.c:apply_textdelta()
>  /* Only compare checksums this file has an entry, and the entry has
>     a checksum.  If there's no entry, it just means the file is
>     created in this update, so there won't be any previously recorded
>     checksum to compare against.  If no checksum, well, for backwards
>     compatibility we assume that no checksum always matches. */
>  if (ent && ent->checksum)
>    {

subversion/libsvn_wc/adm_crawler.c:svn_wc_transmit_text_deltas()
>      /* For backwards compatibility, no checksum means assume a match. */
>     if (ent->checksum)
>       {
>         const char *tb = svn_wc__text_base_path (path, FALSE, pool);
>         unsigned char tb_digest[APR_MD5_DIGESTSIZE];

I think it is dangerous and propose fix version number in which text
base checksums required.
Next question what we should do if no checksum found: display error or
assertion?
See my proposed idea (I haven't add checking for WC format yet):
[[
Make text base checksums mandatory.
* subversion/libsvn_wc/adm_crawler.c
  (svn_wc_transmit_text_deltas): Require checksum if not full text.
* subversion/libsvn_wc/update_editor.c
  (apply_textdelta): Require checksum if text base exists.
]]
Index: subversion/libsvn_wc/adm_crawler.c
===================================================================
--- subversion/libsvn_wc/adm_crawler.c  (revision 16267)
+++ subversion/libsvn_wc/adm_crawler.c  (working copy)
@@ -827,7 +827,16 @@
                 }
             }
         }
+      else
+        {
+          const char *tb = svn_wc__text_base_path (path, FALSE, pool);

+          return svn_error_createf
+            (SVN_ERR_WC_CORRUPT_TEXT_BASE, NULL,
+              _("Missing checksum for '%s'"),
+              svn_path_local_style (tb, pool));
+        }
+
       SVN_ERR (svn_wc__open_text_base (&basefile, path, APR_READ, pool));
     }

Index: subversion/libsvn_wc/update_editor.c
===================================================================
--- subversion/libsvn_wc/update_editor.c        (revision 16267)
+++ subversion/libsvn_wc/update_editor.c        (working copy)
@@ -1666,6 +1666,18 @@
       hb->source = NULL;  /* make sure */
     }

+  if (hb->source != NULL && ent && ent->checksum == NULL)
+    {
+      const char *tb;
+
+      tb = svn_wc__text_base_path (fb->path, FALSE, pool);
+
+      return svn_error_createf
+        (SVN_ERR_WC_CORRUPT_TEXT_BASE, NULL,
+          _("Missing checksum for '%s'"),
+          svn_path_local_style (tb, pool));
+    }
+
   /* Open the text base for writing (this will get us a temporary file).  */
   hb->dest = NULL;
   err = svn_wc__open_text_base (&hb->dest, fb->path,


--
Ivan Zhakov

Re: wc-replacements: text base checksums

Posted by Erik Huelsmann <eh...@gmail.com>.
On 9/26/05, Ivan Zhakov <ch...@gmail.com> wrote:
> Hi!
> I studying working copy checksums for text base and revert base
> behavior. I wonder that checksums is optional for text base.

Since we have upgraded the working copy format number, we can now
require checksums. You'll need to check whether the format is
<whatever-our-current-version-is> or higher. Below this version, we'll
still need to allow for non-existing md5's for backward compatibility.

When the wc-format is upgraded, md5's will have to be calculated for
all files which miss them.

> For
> example if we somewhere forget write checksum for text base, working
> copy would silently apply and commit txt-delta without checking
> checksum:
> subversion/libsvn_wc/update_editor.c:apply_textdelta()
> >  /* Only compare checksums this file has an entry, and the entry has
> >     a checksum.  If there's no entry, it just means the file is
> >     created in this update, so there won't be any previously recorded
> >     checksum to compare against.  If no checksum, well, for backwards
> >     compatibility we assume that no checksum always matches. */
> >  if (ent && ent->checksum)
> >    {
>
> subversion/libsvn_wc/adm_crawler.c:svn_wc_transmit_text_deltas()
> >      /* For backwards compatibility, no checksum means assume a match. */
> >     if (ent->checksum)
> >       {
> >         const char *tb = svn_wc__text_base_path (path, FALSE, pool);
> >         unsigned char tb_digest[APR_MD5_DIGESTSIZE];
>
> I think it is dangerous and propose fix version number in which text
> base checksums required.
> Next question what we should do if no checksum found: display error or
> assertion?

Since no checksum means essentially: 'invalid external input', I'd say
we need to error out. It's bad to sigabort on invalid (external)
input.

(Note: I didn't look at the patch yet.)

bye,


Erik.