Bring main onto the line this project actually ships

main had drifted to a history with no ancestor in common with the tree we
build and deploy. It sat at be63e83 (2026-08-09) with 28 commits; the shipped
line, deploy/soak-2026-08-21, had 383. The two shared nothing -- git
merge-base returns no result at all -- so anyone cloning this repository
landed on a default branch that was neither current nor related to the
running node.

The 28 commits on main were audited before this merge: every one is already
present in the tree by content, including the two that are hardest to see
because they were rewritten afterwards -- the payout reorg-safety work
(PayoutBlockStillOnChain, since evolved to re-check on every retry rather
than once at maturity) and the nostr relay certificate verification
(SSL_set1_host, now with a Windows root-store path). Nothing on main is lost
by this merge.

Recorded as a merge rather than a force-push: main only ever moves forward,
its old history stays reachable as the first parent, and this is revertible.
The tree is taken wholesale from the merged branch, so main's content is now
exactly what was built and tested.

What this brings onto main:

- compactblock: BIP152-style compact block relay ported from upstream 1.2.23,
  primitives and unit test only, with the RAND_bytes failure path closed and
  the reconstruction allocation bounded.
- upstream/main 17741d9: isolated testnet parameters and genesis, optional
  2 MB large pages for RandomX. Mainnet genesis verified byte-identical
  through upstream's BuildGenesisBlock() refactor.
- deploy: every image built on bookworm so binaries run on Debian 12, and the
  container build now honours the RandomX and libsecp256k1 pins it had been
  silently ignoring since before 1.2.15.

Gates at this tree: 15/15 selftests, compactblock's 17 checks, consensus
constants unchanged against e4f898c apart from the pre-existing
MAX_BLOCK_SIZE_GEN, mainnet genesis intact, money-log check ok.

1 files changed, 87 insertions, 6 deletions

diff --git a/src/util.h b/src/util.h
index d72518b..dfe7511 100644
--- a/src/util.h
+++ b/src/util.h
@@ -87,6 +87,15 @@ inline T& REF(const T& val)
 
 extern bool fDebug;
 
+// /debugrelay: the per-message firehose ("relay" and "nostrdiscovery"
+// categories) that even -debug keeps off. See LogAcceptsCategory().
+extern bool fDebugRelay;
+
+// debug.log is rolled to debug.log.old once it grows past this many bytes
+// (set from -maxdebugfilesize, in MB; 0 disables the check). See the roll
+// site in OutputDebugStringF() below.
+extern int64 nMaxDebugFileSize;
+
 // ---------------------------------------------------------------------------
 // Categorized, burst-collapsing logging
 //
@@ -106,15 +115,29 @@ extern bool fDebug;
 //   2. Burst collapse: repeated LogPrint() calls from the *same source line*
 //      only emit one line per DEDUP_WINDOW_SECS; if more calls land inside
 //      that window they're counted instead of reprinted, and the count is
-//      flushed as a single "(xN similar in Ws)" line the moment the window
-//      rolls over. A loop discovering 500 peers by address now produces a
-//      small, bounded number of lines instead of 500 -- log growth is capped
-//      by elapsed time, not by how many similar events happen to fire.
+//      flushed as a single "(xN similar in Ws)" line once the window expires
+//      -- whether or not that call site ever fires again. A loop discovering
+//      500 peers by address now produces a small, bounded number of lines
+//      instead of 500 -- log growth is capped by elapsed time, not by how
+//      many similar events happen to fire.
+//
+// Use LogPrintAlways() instead for anything that reports an irreversible
+// change to money: an obligation voided, a payment sent, a payment that
+// failed. One line standing in for an unknown number of identical events is
+// a tidy report of peer chatter and a false report of a ledger. A reorg
+// voiding 100 entries hits one source line 100 times inside one window;
+// under LogPrint() that read as a single block losing a single 49.49 BTF
+// payment, while 4,950 BTF of obligations were actually destroyed, and the
+// discrepancy was only found by counting entries in pending_payouts.json
+// (issue #3). Bounded log growth is not worth that ambiguity here.
 // ---------------------------------------------------------------------------
 bool LogAcceptsCategory(const char* category);
 void LogPrintDedup(const char* category, const char* file, int line, const std::string& msg);
+void LogPrintNoCollapse(const char* category, const std::string& msg);
 #define LogPrint(category, ...) \
     do { if (LogAcceptsCategory(category)) LogPrintDedup((category), __FILE__, __LINE__, strprintf(__VA_ARGS__)); } while (0)
+#define LogPrintAlways(category, ...) \
+    do { if (LogAcceptsCategory(category)) LogPrintNoCollapse((category), strprintf(__VA_ARGS__)); } while (0)
 
 // Every format string in this project goes through one of the four functions
 // below -- `printf` itself is remapped to OutputDebugStringF just above, and
@@ -145,7 +168,7 @@ void ParseString(const string& str, char c, vector<string>& v);
 string FormatMoney(int64 n, bool fPlus=false);
 bool ParseMoney(const char* pszIn, int64& nRet);
 bool FileExists(const char* psz);
-int GetFilesize(FILE* file);
+int64 GetFilesize(FILE* file);
 uint64 GetRand(uint64 nMax);
 int64 GetTime();
 int64 GetAdjustedTime();
@@ -199,7 +222,7 @@ protected:
 public:
     char* pszFile;
     int nLine;
-    explicit CCriticalSection() { InitializeCriticalSection(&cs); }
+    explicit CCriticalSection() : pszFile(NULL), nLine(0) { InitializeCriticalSection(&cs); }
     ~CCriticalSection() { DeleteCriticalSection(&cs); }
     void Enter() { EnterCriticalSection(&cs); }
     void Leave() { LeaveCriticalSection(&cs); }
@@ -390,7 +413,65 @@ inline int OutputDebugStringF(const char* pszFormat, ...)
             else
                 strDebugFile = "debug.log";
         }
+        // Latched by a failed rename in the roll below, which cannot be
+        // retried usefully and must not be retried endlessly.
+        static bool fRollDisabled = false;
         FILE* fileout = fopen(strDebugFile.c_str(), "a");
+        if (fileout && nMaxDebugFileSize > 0 && !fRollDisabled)
+        {
+            // Roll an overgrown debug.log before adding to it. Nothing in the
+            // tree shrinks this file once the process is running (there is no
+            // ShrinkDebugFile, and even Bitcoin's only ran at startup), so a
+            // node that stays up for months grows the log without bound --
+            // production hit ~2 GiB. The file is opened and closed per line,
+            // so a rename is safe here: the next line simply reopens the
+            // path. One generation is kept as debug.log.old; logrotate
+            // (contrib/logrotate/) handles anything longer-term.
+            int64 nSize = GetFilesize(fileout);
+            if (nSize >= 0 && nSize > nMaxDebugFileSize)
+            {
+                fclose(fileout);
+                string strOldFile = strDebugFile + ".old";
+                remove(strOldFile.c_str()); // rename() won't clobber on Windows
+                bool fRolled = (rename(strDebugFile.c_str(), strOldFile.c_str()) == 0);
+                int nRenameErrno = errno; // before fopen() can overwrite it
+                fileout = fopen(strDebugFile.c_str(), "a");
+                if (fRolled)
+                {
+                    if (fileout)
+                        fprintf(fileout,
+                            "debug.log grew past %lld MB; previous contents moved to debug.log.old (-maxdebugfilesize)\n",
+                            (long long)(nMaxDebugFileSize / 1024 / 1024));
+                }
+                else if (nRenameErrno == ENOENT)
+                {
+                    // The file went away between the size check and the
+                    // rename, so an external rotator (logrotate) got there
+                    // first. Nothing left to roll and nothing wrong: the
+                    // fopen above has already recreated it. Latching on this
+                    // would let a benign race with logrotate switch the cap
+                    // off for the rest of the run.
+                }
+                else
+                {
+                    // The file is still over the cap, so this branch would be
+                    // reached again on the next line and every line after it.
+                    // Unlatched, the complaint about failing to roll becomes
+                    // the runaway growth the roll exists to prevent. Say it
+                    // once and stop trying -- on stderr too, because whatever
+                    // blocked the rename (read-only mount, permissions, EXDEV
+                    // across a bind mount) may equally block the log write.
+                    fRollDisabled = true;
+                    fprintf(stderr,
+                        "cannot roll %s (rename to .old failed: %s); -maxdebugfilesize disabled for this run\n",
+                        strDebugFile.c_str(), strerror(nRenameErrno));
+                    if (fileout)
+                        fprintf(fileout,
+                            "cannot roll debug.log (rename to .old failed: %s); -maxdebugfilesize disabled for this run\n",
+                            strerror(nRenameErrno));
+                }
+            }
+        }
         if (fileout)
         {
             va_list arg_ptr;